No. Supply one image URL and the model separates the foreground from the background.

Is the output actually transparent?

The output is a PNG with an alpha channel. The checkerboard on this page illustrates transparency and is not embedded in the original downloadable result.

Which fields does the API require?

Submit model and input.image_url. The image must be directly accessible over HTTP or HTTPS.

Can I change the aspect ratio, colors or background?

This endpoint focuses on background removal and offers no aspect-ratio, background-color or prompt controls. Use the transparent output in a subsequent editing step.

How should I save the result?

Download it after the task succeeds or copy it to your own storage. Page examples and temporary task outputs have different retention policies.

How do I make the mask?

The mask is a black-and-white image: the white area is erased and the background is filled in, the black area stays unchanged. Paint the white area slightly beyond the object edge so shadows and edge remnants go too.

Does the mask have to match the image size?

The mask must have the same aspect ratio as the image. Exporting it at the image's exact size keeps the erased area in place.

No. Bria Eraser works from the image and the mask only; requests must not include a prompt field.

Should mask_type be manual or automatic?

Use manual (the default) for masks painted with a brush or an editor; use automatic for masks produced by a segmentation tool.

What happens with a transparent input?

Turn on preserve_alpha to keep the input's transparent areas transparent in the result. It is off by default.

What format and size is the output?

One PNG per request at the same resolution as the input image (measured: 1360 × 768 in, 1360 × 768 out). The PNG is RGBA and fully opaque unless preserve transparency is on.

Which input formats are supported?

The image can be JPEG, PNG or WEBP. Both the image and the mask must be publicly reachable http(s) URLs; files uploaded in the Playground are turned into links for you.

Per output image, one image per request. See the live pricing on this page for the current rate.

Can I use a scene description and a scene reference together?

No, pick one. Leave out the reference image when you write an English scene description (prompt), and leave out the prompt when you follow a reference image. Switching mode in the Playground clears the other one.

Does the scene description have to be in English?

Yes. Describe the new scene or background in English, up to 1,000 characters; the Playground asks you to rewrite non-English text.

What does optimize description do?

It expands your scene description automatically and is on by default. It applies only to scene-description mode and is not sent with a reference image.

How can the product be placed?

Preset position (manual_placement, the default) (with either a scene description or a scene reference) puts the product at a chosen spot, bottom center by default; measured 1024×1024 for square input. Keep original position (original) keeps the product where it is in the input photo.

When can I keep the original size?

Only with keep original position: on keeps the input size, off scales to about 1 MP. It is not sent in preset-position mode.

What format and size is the output?

One PNG per request. Preset position (with a scene description or a scene reference) measured 1024 × 1024 for square input; with keep original position and keep original size on, the output keeps the input size (measured: 1168 × 880 in, 1168 × 880 out).

What is the difference between fast and standard mode?

Fast mode (on by default) returns results sooner; turn it off for the standard mode. Both cost the same.

What product images work?

JPEG, PNG or WEBP up to 12 MB, as a publicly reachable http(s) URL. A clear product with complete edges works best.

Per output image, one image per request. See the live pricing on this page for the current rate.

What inputs does Creatify Boreal accept?

A prompt is required; one reference image and one driving audio file are optional. Use the prompt alone or combine all three.

How should I write the prompt?

Use four sections: [VISUAL] for the shot and camera, [SPEECH] for the line, [SOUNDS] for effects and music and [TEXT] for on-screen text. Write None for sections you do not need.

Does it support languages other than English?

English prompts give the steadiest results today; other languages are not guaranteed, so start in English.

How long can a video be?

1 to 20 whole seconds, 10 by default.

Which resolutions are supported?

720p, 1080p and 2K, 720p by default. 2K is 2016 px on the long edge for 16:9 and 9:16 and the 1440p tier for 1:1, 4:3 and 3:4.

How do aspect ratios work?

Choose auto, 16:9, 9:16, 1:1, 4:3 or 3:4. Auto follows the reference image, or 16:9 without one; a named ratio with a reference image centre-crops the image.

How does driving audio work?

Upload narration, dialogue or a soundtrack and the video keeps it. Only the first duration seconds are used, and shorter audio is padded with silence.

What does the reference image do?

It sets the subject's look, outfit and composition, so a product photo or portrait can become a video.

What is the output format?

MP4 (H.264 video + AAC audio, 24fps) with generated speech and sound.

Is there a watermark?

Only an invisible one. Every video carries C2PA metadata and an invisible watermark, with no visible mark on the picture.

By the requested duration and the chosen resolution; see the live price on this page. Failed tasks are refunded in full.

How do I call it through the API?

Submit a task to POST /v1/tasks with model creatify-boreal, then poll the task for the video URL. See the API docs for details.

How do the official name and the HiAPI model ID differ?

DeepSeek's official model name is deepseek-flash. HiAPI's public model ID for this integration remains deepseek-v4.1-flash. Continue using deepseek-v4.1-flash in requests and examples.

What is DeepSeek V4.1 Flash?

DeepSeek V4.1 Flash is a production model for text and image input. According to DeepSeek's official documentation, it supports a 1M-token context window, up to 384K output tokens, and none, low, high, and max reasoning efforts, with high as the default.

Which model ID and endpoint should I use?

Set model to deepseek-v4.1-flash. Use POST /v1/responses or the compatible POST /v1/chat/completions endpoint. Responses uses input; Chat Completions uses messages. Do not mix their request shapes.

How do I enable or disable reasoning?

Responses uses reasoning.effort: none disables reasoning; low, high, and max enable it, with high as the default. Chat Completions uses thinking.type=enabled/disabled with reasoning_effort=low, high, or max when enabled.

How do multi-turn Responses conversations work?

This endpoint is stateless. Send the required conversation history in input on every request. Do not send previous_response_id, conversation, background, or context_management, and omit store.

How does a streamed response end?

Read text increments from response.output_text.delta and final usage from response.completed. response.incomplete and response.failed indicate incomplete generation and failure. Responses does not end with [DONE]; Chat Completions streams still use [DONE].

How are tokens and reasoning billed?

Input, output, and cached input tokens are billed separately. Reasoning tokens count as output. Check usage.input_tokens, usage.output_tokens, and input_tokens_details.cached_tokens against the API log. Refer to the live pricing section for current rates.

Are image input, tool calls, and structured JSON enabled?

DeepSeek's official documentation supports image URLs and base64 image input. Tool calls and structured JSON are not advertised for this integration.

Can I reuse an existing HiAPI API key?

Yes, if it has access to the model and available balance. With the OpenAI SDK, set base_url to https://api.hiapi.ai/v1. Use client.responses.create for Responses or client.chat.completions.create for Chat Completions.

ElevenLabs' low-latency text-to-speech model for voice assistants, notifications, and high-volume voiceover.

Send model (elevenlabs/text-to-speech-turbo-2-5) and input to POST /v1/tasks, then poll with taskId or configure a callback.

Per 1,000 characters of input.text. See the pricing section above for the current rate.

Are previous_text and next_text billed?

No. They only give context for segmented synthesis so prosody joins smoothly.

How are CJK characters counted?

Every character, including each CJK character, letter, digit, punctuation mark, or space, counts once.

What is the maximum text length?

5,000 characters. Longer text is rejected at submission without charge; split long content into segments.

Which voices are available?

67 ElevenLabs preset voices covering narrators, hosts, characters, and support agents. Pass a voice_id as voice.

Can I force the output language?

Yes. Set language_code to a two-letter ISO 639-1 code such as zh, en, or ja; omit it for auto-detection.

Frequently Asked Questions

Quick answers on billing, accounts, API integration, model capabilities, and persistent artifacts.

Categories 1050

Categories

Do I need a prompt?

No. Supply one image URL and the model separates the foreground from the background.

Models & Capabilities 851 Labs Background Remover
Is the output actually transparent?

The output is a PNG with an alpha channel. The checkerboard on this page illustrates transparency and is not embedded in the original downloadable result.

Models & Capabilities 851 Labs Background Remover
Which fields does the API require?

Submit model and input.image_url. The image must be directly accessible over HTTP or HTTPS.

Models & Capabilities 851 Labs Background Remover
Can I change the aspect ratio, colors or background?

This endpoint focuses on background removal and offers no aspect-ratio, background-color or prompt controls. Use the transparent output in a subsequent editing step.

Models & Capabilities 851 Labs Background Remover
How should I save the result?

Download it after the task succeeds or copy it to your own storage. Page examples and temporary task outputs have different retention policies.

Models & Capabilities 851 Labs Background Remover
How do I make the mask?

The mask is a black-and-white image: the white area is erased and the background is filled in, the black area stays unchanged. Paint the white area slightly beyond the object edge so shadows and edge remnants go too.

Models & Capabilities BRIA Eraser
Does the mask have to match the image size?

The mask must have the same aspect ratio as the image. Exporting it at the image's exact size keeps the erased area in place.

Models & Capabilities BRIA Eraser
Do I need a prompt?

No. Bria Eraser works from the image and the mask only; requests must not include a prompt field.

Models & Capabilities BRIA Eraser
Should mask_type be manual or automatic?

Use manual (the default) for masks painted with a brush or an editor; use automatic for masks produced by a segmentation tool.

Models & Capabilities BRIA Eraser
What happens with a transparent input?

Turn on preserve_alpha to keep the input's transparent areas transparent in the result. It is off by default.

Models & Capabilities BRIA Eraser
What format and size is the output?

One PNG per request at the same resolution as the input image (measured: 1360 × 768 in, 1360 × 768 out). The PNG is RGBA and fully opaque unless preserve transparency is on.

Models & Capabilities BRIA Eraser
Which input formats are supported?

The image can be JPEG, PNG or WEBP. Both the image and the mask must be publicly reachable http(s) URLs; files uploaded in the Playground are turned into links for you.

Models & Capabilities BRIA Eraser
How is it billed?

Per output image, one image per request. See the live pricing on this page for the current rate.

Models & Capabilities BRIA Eraser
Can I use a scene description and a scene reference together?

No, pick one. Leave out the reference image when you write an English scene description (prompt), and leave out the prompt when you follow a reference image. Switching mode in the Playground clears the other one.

Models & Capabilities BRIA Product Shot
Does the scene description have to be in English?

Yes. Describe the new scene or background in English, up to 1,000 characters; the Playground asks you to rewrite non-English text.

Models & Capabilities BRIA Product Shot
What does optimize description do?

It expands your scene description automatically and is on by default. It applies only to scene-description mode and is not sent with a reference image.

Models & Capabilities BRIA Product Shot
How can the product be placed?

Preset position (manual_placement, the default) (with either a scene description or a scene reference) puts the product at a chosen spot, bottom center by default; measured 1024×1024 for square input. Keep original position (original) keeps the product where it is in the input photo.

Models & Capabilities BRIA Product Shot
When can I keep the original size?

Only with keep original position: on keeps the input size, off scales to about 1 MP. It is not sent in preset-position mode.

Models & Capabilities BRIA Product Shot
What format and size is the output?

One PNG per request. Preset position (with a scene description or a scene reference) measured 1024 × 1024 for square input; with keep original position and keep original size on, the output keeps the input size (measured: 1168 × 880 in, 1168 × 880 out).

Models & Capabilities BRIA Product Shot
What is the difference between fast and standard mode?

Fast mode (on by default) returns results sooner; turn it off for the standard mode. Both cost the same.

Models & Capabilities BRIA Product Shot
What product images work?

JPEG, PNG or WEBP up to 12 MB, as a publicly reachable http(s) URL. A clear product with complete edges works best.

Models & Capabilities BRIA Product Shot
How is it billed?

Per output image, one image per request. See the live pricing on this page for the current rate.

Models & Capabilities BRIA Product Shot
What inputs does Creatify Boreal accept?

A prompt is required; one reference image and one driving audio file are optional. Use the prompt alone or combine all three.

Models & Capabilities Creatify Boreal
How should I write the prompt?

Use four sections: [VISUAL] for the shot and camera, [SPEECH] for the line, [SOUNDS] for effects and music and [TEXT] for on-screen text. Write None for sections you do not need.

Models & Capabilities Creatify Boreal
Does it support languages other than English?

English prompts give the steadiest results today; other languages are not guaranteed, so start in English.

Models & Capabilities Creatify Boreal
How long can a video be?

1 to 20 whole seconds, 10 by default.

Models & Capabilities Creatify Boreal
Which resolutions are supported?

720p, 1080p and 2K, 720p by default. 2K is 2016 px on the long edge for 16:9 and 9:16 and the 1440p tier for 1:1, 4:3 and 3:4.

Models & Capabilities Creatify Boreal
How do aspect ratios work?

Choose auto, 16:9, 9:16, 1:1, 4:3 or 3:4. Auto follows the reference image, or 16:9 without one; a named ratio with a reference image centre-crops the image.

Models & Capabilities Creatify Boreal
How does driving audio work?

Upload narration, dialogue or a soundtrack and the video keeps it. Only the first duration seconds are used, and shorter audio is padded with silence.

Models & Capabilities Creatify Boreal
What does the reference image do?

It sets the subject's look, outfit and composition, so a product photo or portrait can become a video.

Models & Capabilities Creatify Boreal
What is the output format?

MP4 (H.264 video + AAC audio, 24fps) with generated speech and sound.

Models & Capabilities Creatify Boreal
Is there a watermark?

Only an invisible one. Every video carries C2PA metadata and an invisible watermark, with no visible mark on the picture.

Models & Capabilities Creatify Boreal
How is it billed?

By the requested duration and the chosen resolution; see the live price on this page. Failed tasks are refunded in full.

Models & Capabilities Creatify Boreal
How do I call it through the API?

Submit a task to POST /v1/tasks with model creatify-boreal, then poll the task for the video URL. See the API docs for details.

Models & Capabilities Creatify Boreal
How do the official name and the HiAPI model ID differ?

DeepSeek's official model name is deepseek-flash. HiAPI's public model ID for this integration remains deepseek-v4.1-flash. Continue using deepseek-v4.1-flash in requests and examples.

Models & Capabilities
What is DeepSeek V4.1 Flash?

DeepSeek V4.1 Flash is a production model for text and image input. According to DeepSeek's official documentation, it supports a 1M-token context window, up to 384K output tokens, and none, low, high, and max reasoning efforts, with high as the default.

Models & Capabilities
Which model ID and endpoint should I use?

Set model to deepseek-v4.1-flash. Use POST /v1/responses or the compatible POST /v1/chat/completions endpoint. Responses uses input; Chat Completions uses messages. Do not mix their request shapes.

Models & Capabilities
How do I enable or disable reasoning?

Responses uses reasoning.effort: none disables reasoning; low, high, and max enable it, with high as the default. Chat Completions uses thinking.type=enabled/disabled with reasoning_effort=low, high, or max when enabled.

Models & Capabilities
How do multi-turn Responses conversations work?

This endpoint is stateless. Send the required conversation history in input on every request. Do not send previous_response_id, conversation, background, or context_management, and omit store.

Models & Capabilities
How does a streamed response end?

Read text increments from response.output_text.delta and final usage from response.completed. response.incomplete and response.failed indicate incomplete generation and failure. Responses does not end with [DONE]; Chat Completions streams still use [DONE].

Models & Capabilities
How are tokens and reasoning billed?

Input, output, and cached input tokens are billed separately. Reasoning tokens count as output. Check usage.input_tokens, usage.output_tokens, and input_tokens_details.cached_tokens against the API log. Refer to the live pricing section for current rates.

Models & Capabilities
Are image input, tool calls, and structured JSON enabled?

DeepSeek's official documentation supports image URLs and base64 image input. Tool calls and structured JSON are not advertised for this integration.

Models & Capabilities
Can I reuse an existing HiAPI API key?

Yes, if it has access to the model and available balance. With the OpenAI SDK, set base_url to https://api.hiapi.ai/v1. Use client.responses.create for Responses or client.chat.completions.create for Chat Completions.

Models & Capabilities
What is Turbo 2.5?

ElevenLabs' low-latency text-to-speech model for voice assistants, notifications, and high-volume voiceover.

Models & Capabilities ElevenLabs
How do I call it?

Send model (elevenlabs/text-to-speech-turbo-2-5) and input to POST /v1/tasks, then poll with taskId or configure a callback.

Models & Capabilities ElevenLabs
How is it billed?

Per 1,000 characters of input.text. See the pricing section above for the current rate.

Models & Capabilities ElevenLabs
Are previous_text and next_text billed?

No. They only give context for segmented synthesis so prosody joins smoothly.

Models & Capabilities ElevenLabs
How are CJK characters counted?

Every character, including each CJK character, letter, digit, punctuation mark, or space, counts once.

Models & Capabilities ElevenLabs
What is the maximum text length?

5,000 characters. Longer text is rejected at submission without charge; split long content into segments.

Models & Capabilities ElevenLabs
Which voices are available?

67 ElevenLabs preset voices covering narrators, hosts, characters, and support agents. Pass a voice_id as voice.

Models & Capabilities ElevenLabs
Can I force the output language?

Yes. Set language_code to a two-letter ISO 639-1 code such as zh, en, or ja; omit it for auto-detection.

Models & Capabilities ElevenLabs
How should I tune stability, similarity_boost, and style?

Lower stability is livelier, higher is steadier; similarity_boost controls voice adherence; style adds drama — keep it at 0 for neutral reads.

Models & Capabilities ElevenLabs
Can I get subtitle timestamps?

Yes. Set timestamps=true for word-level timings for subtitles and highlights.

Models & Capabilities ElevenLabs
Turbo 2.5 or Multilingual v2?

Choose Turbo 2.5 for speed and cost (with language_code); choose Multilingual v2 for long-form quality and emotional consistency. Both share the voice library.

Models & Capabilities ElevenLabs
Is voice cloning supported?

The current contract exposes preset voices only; voice cloning and custom voice IDs are not supported.

Models & Capabilities ElevenLabs
Are failed tasks charged?

Failed tasks are not billed; credits reserved at submission are returned automatically.

Models & Capabilities ElevenLabs
What is Multilingual v2?

ElevenLabs' high-quality multilingual text-to-speech model for audiobooks, narration, and brand voiceover.

Models & Capabilities ElevenLabs
What is FASHN Try-On v1.6?

It is a dedicated virtual try-on model. Upload one person photo and one garment image, and it returns a portrait image of the person wearing the garment. No prompt or text description is involved.

Models & Capabilities FASHN Tryon v1.6
Does FASHN Try-On v1.6 need a prompt?

No. It takes a person photo and a garment image only; the try-on result comes from those images plus optional category, mode and garment-photo-type settings.

Models & Capabilities FASHN Tryon v1.6
Do I need to upload both images?

Yes. Both the person photo and the garment image are required; the Playground and API reject a request missing either one.

Models & Capabilities FASHN Tryon v1.6
Which garment categories are supported?

Tops, bottoms and one-pieces, plus automatic detection. Start with auto when the category is unclear, then set it explicitly if detection is wrong.

Models & Capabilities FASHN Tryon v1.6
Which mode should I choose?

performance is faster for batch previews; balanced trades a little speed for quality; quality is slower and intended for final images. You can confirm the composition with balanced, then output the final version with quality.

Models & Capabilities FASHN Tryon v1.6
How do I handle flat-lay, hanging and on-model photos?

Set garment_photo_type to model or flat-lay, or leave it on auto. Flat-lay, hanging, ghost-mannequin and on-model product photos are all supported; an explicit type is usually more stable.

Models & Capabilities FASHN Tryon v1.6
What size and format is the output?

Each request returns one portrait image whose size follows the person photo. Verified runs returned 972 × 1296 from a 3:4 photo and 768 × 1152 from a 2:3 photo. PNG is the default; JPEG is optional. Do not assume a single fixed pixel size.

Models & Capabilities FASHN Tryon v1.6
Why does the same input produce a different result?

Generation has a random element. Pass a fixed seed to reproduce a result, or change the seed for another version. Switching modes also changes how much garment edge and texture detail is resolved.

Models & Capabilities FASHN Tryon v1.6
Will garment details match the reference exactly?

Not pixel for pixel. The model preserves the silhouette, color and main patterns, but details such as necklines, folds and buttons can adapt to the person's pose, category and mode. For important product images, use quality mode and review manually. A front-facing, single-person photo with the body clearly visible works best; group photos and heavy occlusion reduce fit quality.

Models & Capabilities FASHN Tryon v1.6
What edits can FIBO Edit 1.5 do?

Recolors, background swaps, season or style changes, local replacements, and combining several images, such as garment try-ons or placing a product in a new scene.

Models & Capabilities
How many images can I pass?

image_urls accepts 1–4 public image URLs. Order matters: "image 1, image 2" in the instruction follow upload order.

Models & Capabilities
When does aspect_ratio apply?

Only with 2 or more images: 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16 or 16:9, following the first image if omitted. With one image the output keeps that image's ratio.

Models & Capabilities
How do masks work?

mask_url takes a black-and-white image the same size as the source: white is edited, black is kept.

Models & Capabilities
Can I use several images with a mask?

No. With mask_url, image_urls must contain exactly 1 image or the request is rejected; leave the mask out for multi-image composites.

Models & Capabilities
How should I write the instruction?

Say what to change, what it should become and what must stay, e.g. "change only the jacket to deep green, keep everything else". With several images, refer to them as image 1, image 2.

Models & Capabilities
What is the output format?

One PNG image per request.

Models & Capabilities
Why do the same inputs give the same result?

seed defaults to 5555, so the same inputs reproduce the same result; change the seed for another version.

Models & Capabilities
Which model should I use without a source image?

Generate one with FIBO Gen 1.5 text-to-image first, then edit it with FIBO Edit 1.5.

Models & Capabilities
What are the image requirements?

Publicly accessible http(s) image URLs; you can also upload images directly in the Playground.

Models & Capabilities
Is the task synchronous?

No, it is asynchronous. Submitting returns a taskId right away; poll the task or use a callback to get the result.

Models & Capabilities
How is it billed?

Per output image, one per request; the number of input images and masks do not affect the price. See the current price at the top of this page and on the pricing page.

Models & Capabilities
What is FIBO Gen 1.5 good for?

Photoreal product shots, portraits, food, architecture, and signs or packaging with text.

Models & Capabilities
Which aspect ratios are supported?

aspect_ratio accepts 1:1 (default), 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16 and 16:9.

Models & Capabilities
How large is the output?

One PNG of about 1 megapixel per request; width and height follow the ratio, for example 1024 × 1024 at 1:1.

Models & Capabilities
Can I get a higher resolution?

Only the roughly 1 MP tier is available for now. For a larger image, pass the result to an upscaling model.

Models & Capabilities
Why does the same prompt give the same image?

seed defaults to 5555, and the same seed and prompt give the same image. Change the seed for a different result.

Models & Capabilities
Is in-image text accurate?

Put the words in quotes and the model aims to render them exactly; short signs and labels work best.

Models & Capabilities
How many images per request?

One image per request. Submit several tasks, or call again with different seeds.

Models & Capabilities
Do I need a reference image?

No. It is a text-to-image model; to change an existing image, use FIBO Edit 1.5.

Models & Capabilities
Is there a prompt length limit?

prompt accepts 1–10000 characters.

Models & Capabilities
Can I keep editing the result?

Yes. Pass the output URL to fibo-edit-1.5/image-to-image and describe the change in one instruction.

Models & Capabilities
How is it billed?

Per output image, one per request. See the current price at the top of this page and on the pricing page.

Models & Capabilities
What is FLUX.1 Fill Pro?

FLUX.1 Fill Pro is a mask-driven image editing model: upload a source image, a same-size black-and-white mask and a prompt, and only the white area is repainted while the black area stays unchanged.

Models & Capabilities FLUX.1 Fill Pro
What do black and white mean in the mask?

Black keeps the source and white is repainted by the prompt. The mask must match the source image dimensions.

Models & Capabilities FLUX.1 Fill Pro
Are both image and mask required?

Yes. The current public contract requires a source image and a black-and-white mask.

Models & Capabilities FLUX.1 Fill Pro
What should the prompt describe?

Describe only the new content inside the white mask and state what must be preserved; do not repeat the area outside the mask.

Models & Capabilities FLUX.1 Fill Pro
How should I choose steps and guidance?

steps ranges from 15 to 50 with a default of 50; higher values add detail. guidance ranges from 1.5 to 100 with a default of 60; higher values follow the prompt more closely.

Models & Capabilities FLUX.1 Fill Pro
Does the output keep the source size?

The output keeps the source aspect ratio. Actual pixels follow the task result; this 1600×900 source produced a 1440×810 image.

Models & Capabilities FLUX.1 Fill Pro
Which input and output formats are supported?

The source and mask accept JPEG, PNG, GIF and WebP; choose jpg or png for the output.

Models & Capabilities FLUX.1 Fill Pro
How can I make results more stable?

Fix seed and reuse the same source, mask and prompt. Changing steps or guidance changes detail and adherence.

Models & Capabilities FLUX.1 Fill Pro
How is it billed?

Per output image, one image per request. See the live price on this page for the current rate.

Models & Capabilities FLUX.1 Fill Pro
What is FLUX.2 [klein] 4B best for?

It fits batch drafts, product concepts, social assets, and cost-sensitive iteration. Move to 9B when realism, typography, and fine detail matter more.

Models & Capabilities FLUX.2 Klein 4B
How should I choose between 0.25 and 4 MP?

Start with 0.25 or 0.5 MP for drafts, use 1 MP for typical web assets, and select 2 or 4 MP for larger output. Price changes by megapixel tier.

Models & Capabilities FLUX.2 Klein 4B
Which aspect ratios are supported?

1:1, 16:9, 9:16, 3:2, 2:3, 4:3, 3:4, 5:4, 4:5, 21:9, and 9:21.

Models & Capabilities FLUX.2 Klein 4B
How can I reproduce a composition direction?

Keep seed, prompt, ratio, megapixel tier, and output settings fixed while changing one variable at a time.

Models & Capabilities FLUX.2 Klein 4B
What belongs in the minimal API request?

POST /v1/tasks with model and input.prompt. Other settings may use defaults.

Models & Capabilities FLUX.2 Klein 4B
How many references does FLUX.2 [klein] 9B Image to Image accept?

It currently accepts one JPEG, PNG, or WebP reference. image_urls must contain exactly one public image URL.

Models & Capabilities FLUX.2 Klein 9B
Which editing tasks is it best for?

It is suited to product cleanup, background replacement, in-image text rewriting, color changes, and commercial edits that preserve subject and camera.

Models & Capabilities FLUX.2 Klein 9B
How do I set inference steps?

num_inference_steps accepts integers from 4 to 8 and defaults to 4. The current price does not change with the step count.

Models & Capabilities FLUX.2 Klein 9B
Which ratios and formats are supported?

Ratios: 1:1, 4:3, 3:4, 16:9, and 9:16. Formats: JPEG, PNG, and WebP.

Models & Capabilities FLUX.2 Klein 9B
How do I submit the reference through the API?

Pass one public image URL in input.image_urls when posting to /v1/tasks, together with the prompt. Ratio, seed, steps, and format are optional.

Models & Capabilities FLUX.2 Klein 9B
What is FLUX.2 [klein] 9B Text to Image best for?

It is suited to fast commercial visuals, product images, English-text posters, and magazine covers. This route does not accept reference images.

Models & Capabilities FLUX.2 Klein 9B
How can I reproduce or compare outputs?

Fix the seed and keep prompt, ratio, format, and inference steps constant while changing only the variable you want to compare.

Models & Capabilities FLUX.2 Klein 9B
Which fields are required in the minimal API request?

POST /v1/tasks with model and input.prompt. Ratio, inference steps, and output format may use their defaults.

Models & Capabilities FLUX.2 Klein 9B
What is FLUX.2 [pro]?

FLUX.2 [pro] is a text-to-image model from Black Forest Labs. HiAPI exposes it through the unified asynchronous task endpoint for workflows such as product imagery, posters, brand visuals, and compositions with detailed written constraints.

Models & Capabilities
How much does FLUX.2 [pro] cost?

FLUX.2 [pro] is priced by output-resolution tier. Preset aspect ratios do not change the price within a tier. See the live page price and task record for the final charge.

Models & Capabilities
What resolutions does FLUX.2 [pro] support?

The current HiAPI endpoint supports 0.5 MP, 1 MP, 2 MP, and 4 MP tiers. With the custom aspect ratio, width and height can be set from 256 to 2048 and are rounded to multiples of 16; the resolution field is ignored in custom mode.

Models & Capabilities
How do I use the FLUX.2 [pro] API?

Send model=flux-2/text-to-image and input.prompt to POST /v1/tasks. The asynchronous response returns a taskId. In production, configure callback.url for terminal notifications or query the task endpoint for the result. The separate API documentation contains the complete parameter and request reference.

Models & Capabilities
Can I try FLUX.2 [pro] online?

Yes. Use the Playground on this page to test a prompt, aspect ratio, resolution, and output format, then use the same model ID when you are ready to integrate the API.

Models & Capabilities
Does FLUX.2 [pro] support image editing?

The flux-2/text-to-image endpoint on this page is for text-to-image generation. For reference-image editing or composition, use the separate flux-2/image-to-image model instead of adding image fields to this request.

Models & Capabilities
Can I use FLUX.2 [pro] commercially?

Commercial use depends on the current Black Forest Labs terms, the HiAPI terms, and the content and jurisdiction involved. Review the latest licensing and compliance requirements before production use; this page is not legal or licensing advice.

Models & Capabilities
What is the difference between FLUX.2 [pro] and FLUX.2 [klein]?

FLUX.2 [pro] is the high-fidelity production option with a published HiAPI model ID, parameter contract, and four resolution-based prices. FLUX.2 [klein] is positioned for efficiency and faster iteration, but this repository does not currently contain a verifiable HiAPI Klein endpoint, price, or specification, so this page avoids publishing unsupported numbers or a dead link.

Models & Capabilities
What inputs does FLUX 3 Max Image to Video need?

prompt and image_url (the first frame, PNG/JPEG/WebP) are required. Optional: end_image_url as the last frame, plus duration, resolution, aspect_ratio, generate_audio and reasoning_level.

Models & Capabilities FLUX 3 Max
Can I set the video length?

Yes. duration accepts whole seconds from 5 to 20, 5 by default. Image to video shares the same per-second pricing as text to video, and a 5-second check shows whether the source stays stable.

Models & Capabilities FLUX 3 Max
What decides the aspect ratio?

aspect_ratio defaults to auto, which follows the first frame's ratio. You can also set 21:9, 2:1, 16:9, 4:3, 1:1, 3:4 or 9:16. If a manual ratio differs from the image, the frame may be cropped.

Models & Capabilities FLUX 3 Max
How do first and last frames work?

Pass the first frame in image_url and the last in end_image_url; the video moves from one to the other. The prompt only needs to describe the change in between, and the closer the two frames match in framing and light, the smoother the transition.

Models & Capabilities FLUX 3 Max
Does adding a last frame cost more?

No. Image to video is billed per output second by resolution, with or without end_image_url, at the same rate as text to video. Only the image count changes, not the per-second price.

Models & Capabilities FLUX 3 Max
What are the first-frame image requirements?

Use a publicly reachable PNG, JPEG or WebP. The Playground accepts a local upload and converts it to a URL. Crop the image to the target ratio first so auto matching does not trim important elements.

Models & Capabilities FLUX 3 Max
Can it generate sound?

Yes. generate_audio is on by default and adds sound that follows the picture; turn it off for a silent master and add your own music or narration later. The setting does not affect the price.

Models & Capabilities FLUX 3 Max
How is FLUX 3 Max Image to Video billed?

Per output second, with the rate set by resolution, so the total is the per-second rate times the duration. Failed tasks are not charged, and a 720p draft is cheaper while still showing whether the source holds.

Models & Capabilities FLUX 3 Max
How do I call it through the API?

Send an asynchronous task to POST /v1/tasks with model set to flux-3-max/image-to-video, prompt and image_url in input, and end_image_url or other settings as needed. The call returns a taskId; get the result through callback.url or GET /v1/tasks/:id.

Models & Capabilities FLUX 3 Max
Image to video or text to video?

Use image to video when you already have the picture, product shot or portrait you want to animate. Use text to video when the whole scene comes from words or you need to choose the ratio freely. Pricing and parameters are otherwise aligned.

Models & Capabilities FLUX 3 Max
What is the keyframes input?

keyframes is an ordered array of 1 to 10 items, each with an image_url and a frame_index. The model connects the keyframes in frame order and generates the motion between them.

Models & Capabilities FLUX 3 Max
How do I set frame_index?

frame_index is the frame number on a 24fps timeline, starting at 0. The maximum is duration × 24; for a 5-second clip that is 120. Every frame index must be unique.

Models & Capabilities FLUX 3 Max
How many keyframes do I need?

At least 1 and at most 10. Two or more create a clear start and end or a rhythm change; a single keyframe only pins one moment and does not produce a real transition.

Models & Capabilities FLUX 3 Max
How do duration and keyframes work together?

Set duration first (5 to 20 seconds), then convert to frames at 24fps: 5 seconds covers 0–120, 10 seconds 0–240 and 20 seconds 0–480. Space the keyframes evenly or concentrate them in one section.

Models & Capabilities FLUX 3 Max
How is this different from first and last frames?

First and last frames only fix the ends. keyframes can place several checkpoints in between, which suits multi-stage shots such as unboxing, plant growth or a day-to-night change. More anchors give finer timing control.

Models & Capabilities FLUX 3 Max
How do I choose the format?

resolution is 720p or 1080p, 720p by default. aspect_ratio defaults to auto and also accepts 21:9, 2:1, 16:9, 4:3, 1:1, 3:4 or 9:16. Keep every keyframe in the same aspect ratio.

Models & Capabilities FLUX 3 Max
How is FLUX 3 Max Keyframes to Video billed?

Per output second, with the rate set by resolution, the same as text to video and image to video. Anchor count does not change the price, and failed tasks are not charged. Validate the timeline with a 5-second draft first.

Models & Capabilities FLUX 3 Max
Which lengths does FLUX 3 Max Text to Video support?

duration accepts any whole number of seconds from 5 to 20 in one-second steps, 5 by default. Longer clips cost more because billing is per second, and a 20-second render uses four times the seconds of a 5-second draft.

Models & Capabilities FLUX 3 Max
How do I choose resolution and aspect ratio?

resolution is 720p or 1080p, 720p by default. aspect_ratio defaults to auto and can also be set to 21:9, 2:1, 16:9, 4:3, 1:1, 3:4 or 9:16.

Models & Capabilities FLUX 3 Max
Does the video have sound?

generate_audio is on by default and adds sound that follows the picture. Turn it off for a silent clip, and use the switch per request instead of baking sound assumptions into the prompt. The setting does not change the price.

Models & Capabilities FLUX 3 Max
What is reasoning_level?

It controls how thoroughly the model interprets the prompt and plans the shot. Choose low or high; high is the default. Use high for complex scenes and low for simple shots.

Models & Capabilities FLUX 3 Max
How should I write a prompt?

Lead with the subject and setting, then describe action, camera direction and speed in time order, and finish with light and mood. Keep one main action per clip and say no cuts for a continuous shot.

Models & Capabilities FLUX 3 Max
Is text inside the picture reliable?

No. The model may invent lettering and cannot guarantee spelling, so add signs, slogans and captions in post and never rely on it for a brand name or legal copy.

Models & Capabilities FLUX 3 Max
Does it work for vertical short video?

Yes. Pick 9:16 in the request instead of cropping a landscape clip. 1:1 and 3:4 suit feeds, and 21:9 suits openers and widescreen content. Keep the subject centered so platform UI does not cover it.

Models & Capabilities FLUX 3 Max
How is FLUX 3 Max Text to Video billed?

Per second of output, with the rate set by resolution, so the total is the per-second rate times the duration. Failed tasks are not charged, and a 5-second draft is the cheapest way to compare two prompts.

Models & Capabilities FLUX 3 Max
How is it different from image to video?

Text to video starts from words alone, so the prompt decides framing and composition. When you already have the opening frame, use flux-3-max/image-to-video instead. Both share the same duration range and per-second rate.

Models & Capabilities FLUX 3 Max
Which FLUX 3 workflows are supported?

The same model ID supports text-to-video, 1 to 10 ordered keyframe images, and MP4 source-video continuation. Do not combine images with start_video.

Models & Capabilities
What is the difference between Draft and Full?

Draft is a lower-cost 720p preview tier for checking composition, pacing, and motion. Full supports 720p and 1080p for final output.

Models & Capabilities
How do FLUX 3 keyframes work?

One image defines the opening, two images define the first and last frames, and 3 to 10 ordered images form a storyboard. The prompt controls motion, camera, and sound.

Models & Capabilities
What are the video continuation limits?

start_video accepts an MP4 up to 50MB and 15 seconds. Do not submit images in the same request. Select an output duration from 5 to 20 seconds.

Models & Capabilities
Can FLUX 3 generate audio?

Yes. generate_audio is enabled by default for synchronized ambience, sound effects, or dialogue. Disable it for video-only output.

Models & Capabilities
How is FLUX 3 priced?

Pricing depends on output duration, resolution, Draft mode, and whether source-video continuation is used. The page lists all six per-second tiers.

Models & Capabilities
Can one entry generate and edit?

Yes. Generate from text without references, or provide 1–10 images to edit around existing material.

Models & Capabilities FLUX 3 Image
How should I choose among the five tiers?

Start with 768 or 1K for a direction, consider 1.5K or 2K for more detail, and 4K when a larger file or cropping room is useful. Tiers do not specify fixed edges; inspect the returned dimensions.

Models & Capabilities FLUX 3 Image
How should I assign reference roles?

Assign the subject, setting and palette or style to specific references in order, including anything that should stay out.

Models & Capabilities FLUX 3 Image
How many images does each task return?

Each task returns one image. Adjust the brief and submit a separate task for another direction.

Models & Capabilities FLUX 3 Image
Can I choose the generated file format?

This entry has no file-format selector. Work with the returned file or convert it in an image tool after downloading.

Models & Capabilities FLUX 3 Image
When is search assistance useful?

Enable it when web or image searches could add visual context before generation. Turn it off for a direction based only on your own brief and references.

Models & Capabilities FLUX 3 Image
How should I keep images long term?

Generated image links have an expiry date. Download files you need to keep, or select permanent storage when creating the task. Existing images can also be re-stored by following the storage management guide.

Models & Capabilities FLUX 3 Image
Can I continue editing a result?

Save the result and supply it as a reference in a new task. Tasks are independent, so describe what to retain and what this revision should change.

Models & Capabilities FLUX 3 Image
How can I make lettering easier to inspect?

Put a short heading in quotes and describe its position, hierarchy and surrounding space. Check the lettering before use and finish typography in a design tool if needed.

Models & Capabilities FLUX 3 Image
When should I leave framing automatic?

Use automatic framing when the placement is flexible. Choose a ratio for a fixed banner, portrait poster or square card, then check the downloaded dimensions.

Models & Capabilities FLUX 3 Image
Where can I check the cost before generating?

Choose your parameters and check the Playground estimate and live pricing. Task records show the completed image and its charge.

Models & Capabilities FLUX 3 Image
Why can the same brief produce a different picture?

Composition and details can vary between results. Use a selected picture as a reference and specify preservation requirements when continuity matters.

Models & Capabilities FLUX 3 Image
How do I choose generation or editing?

Choose generation to start from a written brief, or editing with references to build on an existing subject or layout.

Models & Capabilities Muse Image 1.0
How many images can I get per task?

Each task returns one image with no count selector. Submit a new task with an adjusted prompt for another composition.

Models & Capabilities Muse Image 1.0
Can I select 1K, 2K or 4K?

This entry has no resolution tiers. Choose framing; the model determines dimensions, typically around 2.5MP. Inspect the downloaded file.

Models & Capabilities Muse Image 1.0
Which file formats can I choose?

Choose JPEG, PNG or WebP; the default is WebP. Select a format for your web or design workflow.

Models & Capabilities Muse Image 1.0
How can I leave room for a cover title?

Describe subject placement and which side needs whitespace, then choose framing for the final cover. Avoid relying entirely on a later crop.

Models & Capabilities Muse Image 1.0
What should a product brief start with?

Start with the subject, materials and placement, then add light, background and a few props. State short label text separately and inspect the silhouette and lettering.

Models & Capabilities Muse Image 1.0
How should I keep images long term?

Generated image links have an expiry date. Download files you need to keep, or select permanent storage when creating the task. Existing images can also be re-stored by following the storage management guide.

Models & Capabilities Muse Image 1.0
Can I continue editing a result?

Save the result and supply it as a reference in a new task. Tasks are independent, so describe what to retain and what this revision should change.

Models & Capabilities Muse Image 1.0
How can I make lettering easier to inspect?

Put a short heading in quotes and describe its position, hierarchy and surrounding space. Check the lettering before use and finish typography in a design tool if needed.

Models & Capabilities Muse Image 1.0
When should I leave framing automatic?

Use automatic framing when the placement is flexible. Choose a ratio for a fixed banner, portrait poster or square card, then check the downloaded dimensions.

Models & Capabilities Muse Image 1.0
Where can I check the cost before generating?

Choose your parameters and check the Playground estimate and live pricing. Task records show the completed image and its charge.

Models & Capabilities Muse Image 1.0
Why can the same brief produce a different picture?

Composition and details can vary between results. Use a selected picture as a reference and specify preservation requirements when continuity matters.

Models & Capabilities Muse Image 1.0
Are references required?

Provide 1–10 references. Choose text-to-image to start without an existing picture.

Models & Capabilities Muse Image 1.0
How many edited images are returned?

Each task returns one edited image with no count selector. Submit separate tasks for different revisions.

Models & Capabilities Muse Image 1.0
How do I preserve a subject while changing its setting?

List the recognizable subject cues to retain, then describe the new setting, light and placement. Inspect scale, occlusion and edges.

Models & Capabilities Muse Image 1.0
Will every reference appear in the result?

No. References can guide the subject, layout, palette or style. Assign their roles and specify what should stay out.

Models & Capabilities Muse Image 1.0
How can I replace old lettering?

Specify old wording to remove and new text to include, then describe placement, hierarchy and whitespace. Check the text and layout before use.

Models & Capabilities Muse Image 1.0
How should I choose framing and file format?

Choose landscape, portrait or square framing for the placement, then JPEG, PNG or WebP. There are no resolution tiers; inspect the generated dimensions.

Models & Capabilities Muse Image 1.0
How do I start with text, images or a video?

Start with a prompt for an idea, reference images for a product or character's appearance, or a first frame for the opening, with an optional last frame to guide the ending. Upload a reference video to change an existing shot's mood. All these approaches use the same model.

Models & Capabilities
Which inputs cannot be combined?

first_frame_url cannot be sent together with image_urls or video_list, and last_frame_url only works alongside first_frame_url. image_urls can be combined with video_list, but all media share 7 input slots and one video takes 2 of them, so a request with a video can carry at most 5 images.

Models & Capabilities
How do I get the length I want when reworking video?

With a reference video, output length follows that clip. Upload a clip that matches your target length; do not rely on duration or start and ends to trim the output. Supply at most one reference video, up to 30 seconds and 100MB. The request still requires duration with a value of 4, 6, 8 or 10.

Models & Capabilities
How do I choose the format and resolution?

Use 16:9 for product pages and landscape films, or 9:16 for mobile shorts. Check composition and motion at 360p or 720p, then choose 1080p or 4k for your display needs.

Models & Capabilities
Can I add ambience to the scene?

The output includes ambient sound. Describe rain on a window, quiet lake water or café background sounds in your prompt so audio is part of the scene. Replace the track in editing if you need your own audio.

Models & Capabilities
What are the image requirements?

Each image can be up to 20MB and must be a publicly reachable URL; files uploaded in the Playground are converted to links automatically. First and last frames work best when they share the same framing and aspect ratio.

Models & Capabilities
Where can I check current costs?

See the live pricing page. The Playground shows an estimate for your selected settings before you submit.

Models & Capabilities
How do I get the result?

Create an async task with POST /v1/tasks and read data.taskId. Receive the terminal status through callback.url or poll GET /v1/tasks/{taskId}; when the status is success, download the video from output[].url.

Models & Capabilities
What inputs does Google Virtual Try-On need?

One person photo and one garment image, both required. No prompt is needed.

Models & Capabilities Google Virtual Try-On
What kind of person photo works best?

A single person with the body clearly visible and little occlusion; front-facing or slightly angled standing poses are the most stable.

Models & Capabilities Google Virtual Try-On
What should the garment image look like?

Flat-lay or hanging product photos both work; a complete garment on a clean background with no other people is best.

Models & Capabilities Google Virtual Try-On
How many images can one request return?

One; num_images is fixed at 1. For more options, submit the same input pair a few times; the results differ slightly.

Models & Capabilities Google Virtual Try-On
Which garment types work?

Common tops, jackets and dresses are good candidates; complex accessories or layered outfits may be less consistent.

Models & Capabilities Google Virtual Try-On
Does it change the face or background?

Face, hair, pose and background largely stay as in the original photo; the change centres on the garment.

Models & Capabilities Google Virtual Try-On
What is the output format?

PNG images, sized by the model based on the inputs.

Models & Capabilities Google Virtual Try-On
Can I use photos of real people?

Yes, as long as you have the right to use that person's likeness. The people shown on this page are generated.

Models & Capabilities Google Virtual Try-On
Why does a request sometimes return no image?

The safety check may filter the result; the request is still billed as one image.

Models & Capabilities Google Virtual Try-On
What URLs are accepted?

Publicly reachable http(s) links; in the Playground you can simply upload the files.

Models & Capabilities Google Virtual Try-On
How is it billed?

Per requested image; see the live price on this page. Failed tasks are refunded in full.

Models & Capabilities Google Virtual Try-On
How do I call it through the API?

Submit a task to POST /v1/tasks with model google-virtual-try-on, person_image_url and product_image_url (num_images can be omitted; it is fixed at 1), then poll the task for image URLs. See the API docs for details.

Models & Capabilities Google Virtual Try-On
What is GPT-6.1 Sol suited for?

GPT-6.1 Sol is an OpenAI text model for coding, analysis, research, and tasks that benefit from multistep reasoning. This page describes the streaming Responses text integration; image, audio, and video input or output are outside this integration.

Models & Capabilities
Which model ID and endpoint should I use?

Use model ID gpt-6.1-sol with POST /v1/responses. The request must use an input array, stream=true, and store=false.

Models & Capabilities
Which reasoning.effort values are supported?

The current HiAPI integration supports low, medium, high, xhigh, and max. Examples and Playground default to medium; send reasoning.effort explicitly when you need a fixed setting. Other values are not part of this model contract.

Models & Capabilities
Why are temperature, top_p, and max_output_tokens unavailable?

The current GPT-6.1 Sol integration contract does not expose these fields. Omit them and use reasoning.effort when you need to change reasoning depth. Do not copy sampling parameters from another model into this request.

Models & Capabilities
How should I process the stream?

Read SSE events by type and accumulate response.output_text.delta. Finish on response.completed, response.incomplete, or response.failed. Do not parse the whole stream as one JSON object or rely only on output in the completion event.

Models & Capabilities
How do I continue a conversation with store=false?

Keep the required user messages, assistant items, and tool items in your client, then replay them in the next input array. Replayed history counts as input again; use usage details to determine cache reads and writes.

Models & Capabilities
How are input, cache, and output tokens billed?

Input, cache reads, cache writes, and output use the live token rates shown on the model page. Count a cache read or write only when usage reports cached_tokens or cache_write_tokens; both are included in the input total and must not be added twice. reasoning_tokens are already included in output_tokens and must not be added again; settled charges are recorded in account usage logs.

Models & Capabilities
How does pricing work above 272K input tokens?

The tier uses total input tokens for the request, including cache reads and writes reported in usage. Up to 272,000 input tokens uses the standard tier; from 272,001, the long-context rates apply to the whole request, including input, output, and cache, rather than only the excess.

Models & Capabilities
What should I check when a request fails or has no text?

Check the API key, model, Responses endpoint, input array, stream=true, and store=false first. For missing text, verify that the client handles SSE deltas; for cost questions, compare input_tokens, cached_tokens, cache_write_tokens, and output_tokens in usage.

Models & Capabilities
What is GPT-6 Astra suited for?

GPT-6 Astra is an OpenAI model for complex reasoning, coding, research, and document work, including code review and multistep analysis. This integration covers text input and text output; check the relevant integration reference before using other input or output formats.

Models & Capabilities
Which model ID, endpoint, and SDK settings should I use?

Use model gpt-6-astra and POST /v1/responses. Set your OpenAI SDK base URL to https://api.hiapi.ai/v1, supply a HiAPI API key, and use the Responses client. The same key can access models enabled for your account. Media models using /v1/tasks require a different request structure.

Models & Capabilities
Why do the examples use an input array, stream=true, and store=false?

This integration uses a message array in input, streams events with stream=true, and leaves conversation storage to your client with store=false. Adapt Chat Completions messages and choices handling to Responses, and do not parse an entire SSE response as a single JSON object.

Models & Capabilities
How should I choose a reasoning effort?

Set reasoning.effort to low, medium, high, xhigh, or max. Examples and Playground start with medium; try low for simpler requests and compare higher efforts on difficult work. Higher effort may increase latency and output tokens. Compare results and cost on representative tasks, and send the desired effort explicitly in API requests.

Models & Capabilities
Can I turn reasoning off or adjust sampling parameters?

GPT-6 Astra does not support reasoning.effort=none. This integration does not expose temperature, top_p, or max_output_tokens controls. Follow this model's examples and check the exact endpoint contract when a parameter is rejected.

Models & Capabilities
What do the context and maximum-output limits mean?

The published model limits are a 1,050,000-token context window and 128,000 output tokens. Input and generated content share the context budget, and reasoning consumes output budget. Maximum output is a specification limit, not the expected response length or an adjustable max_output_tokens control in this integration.

Models & Capabilities
Why can response.completed contain an empty output array?

Content may already have arrived through response.output_text.delta or response.output_item.done. Accumulate those events and read status and usage at completion instead of requiring the final event to repeat every output item. Handle response.failed and response.incomplete separately.

Models & Capabilities
How do I continue a conversation, and is history billed again?

With store=false, replay the user messages, assistant items, and tool results needed for the next turn in input. Replayed history counts as input again, with any reported cache hits billed at the cache-read rate. Keep the context needed for the task rather than unrelated history.

Models & Capabilities
How are input, output, cache reads, and cache writes charged?

Each category uses the rate shown in live pricing. Total input includes cache reads and writes; subtract both when calculating ordinary input to avoid charging them twice. Reasoning tokens are already included in output. Page estimates are informational; account usage logs record settled charges.

Models & Capabilities
When does long-context pricing apply?

The threshold uses total input, including cache reads and writes. Exactly 272,000 input tokens stays in the standard band; 272,001 selects the long-context band for the entire request, including output and cache usage. It is not a surcharge only on the excess. Expand the pricing details to compare both bands.

Models & Capabilities
Do repeated prompts always receive a cache discount?

No. Cache writes and hits depend on server policy, reusable prefixes, and retention. Use cached_tokens and cache_write_tokens from usage. Zero means no usage was reported for that category; repeating a prompt alone is not evidence of a cache discount.

Models & Capabilities
How do I use function calls and structured JSON?

Declare functions in tools. Your application executes a returned function_call and sends function_call_output with the same call_id and required context. Use text.format.type=json_object for JSON, or json_schema for field constraints. Assemble streamed JSON before parsing and validating it; see the documented request examples.

Models & Capabilities
What should I check when a request fails or its cost looks wrong?

Check the API key, exact model ID, endpoint, and request fields first. For missing text, verify SSE handling rather than reading only final output. For pricing questions, compare input, cache, output, and the selected context band against usage. Provide the request ID, error code, and time for support, without disclosing your API key.

Models & Capabilities
Are source images required?

Yes. image_urls is required, 1 to 16 images. Use gpt-image-2.5-flare/text-to-image if you only have a prompt.

Models & Capabilities
What is this edit route for?

Keep the subject, then change material or setting. State what must change and what must stay.

Models & Capabilities
How does it differ from Sunburst image-to-image?

Flare is faster for everyday restyling. Sunburst is finer for industrial space and material finish.

Models & Capabilities
How do I choose a ratio?

Pick one of the thirteen ratios, or omit aspect_ratio to use auto.

Models & Capabilities
Which API fields are required?

Submit model, input.prompt, and input.image_urls. aspect_ratio and resolution may be omitted.

Models & Capabilities
What is this route for?

Graphic collage, poster drafts, and everyday stills. Each request returns one image.

Models & Capabilities
How does it differ from Sunburst text-to-image?

Flare is faster for daily output. Sunburst is finer for industrial space, materials, and geometry.

Models & Capabilities
Which ratios are available?

auto plus 1:1, 3:2, 2:3, 4:3, 3:4, 16:9, 9:16, 21:9, 27:16, 16:27, 9:8, and 8:9.

Models & Capabilities
Can I upload references?

Not on this route. Use gpt-image-2.5-flare/image-to-image.

Models & Capabilities
Are source images required?

Yes. image_urls is required, 1 to 16 images. Use gpt-image-2.5-sunburst/text-to-image if you only have a prompt.

Models & Capabilities
What is this edit route for?

Keep industrial structure, then change material, temperature, or surface finish.

Models & Capabilities
How does it differ from Flare image-to-image?

Sunburst is finer for industrial space and materials. Use Flare first for everyday collage restyling.

Models & Capabilities
What is this route for?

Industrial space, materials, and geometry that need finer finish.

Models & Capabilities
How does it differ from Flare text-to-image?

Sunburst is finer and slower. Use Flare first for everyday collage and poster drafts.

Models & Capabilities
Can I upload references?

Not on this route. Use gpt-image-2.5-sunburst/image-to-image.

Models & Capabilities
What is GPT Image 2.5 Flare designed for?

Flare focuses on speed and high-quality everyday generation for creator tools, ad drafts, visual prototypes and frequent image requests. It also supports editing and reference fidelity.

Models & Capabilities
Does Flare only generate from text?

It also accepts images. A reference can supply a person, product or style for a new creation or edit. The pixel-art rider on this page uses an image as a style reference.

Models & Capabilities
Can Flare produce photographic images?

Yes. The sailor and poolside examples show photographic treatments of people, materials and ambient light. Camera and lighting descriptions guide appearance.

Models & Capabilities
Can Flare create an advertisement with a headline?

The requested wording can be described alongside the visual brief. The clothing-campaign example combines people, outfits and a headline in one image. Headlines, colors and styling can be refined in subsequent versions.

Models & Capabilities
Can Flare create comics and recurring characters?

The page shows a four-panel pet story and a forest character continued into another scene. Flare can arrange sequential panels in one image or use a previous output as a reference for the next creation.

Models & Capabilities
What kinds of interface concepts can Flare create?

Flare can create visual concepts for mobile apps, product listings and content pages. These images present layout, color, navigation and information hierarchy for proposals and design discussions.

Models & Capabilities
Can a style reference be used with a different subject?

Yes. The pixel-art example uses a game-screen reference to create a motorcycle rider. The prompt can specify which stylistic elements to borrow and what the new image should contain.

Models & Capabilities
Is Flare an option for an existing GPT Image 2 workflow?

Yes. Flare supports text-to-image generation and reference-based editing with a focus on lower generation latency, making it relevant to content tools, design applications and product displays with frequent image updates.

Models & Capabilities
When should I consider Sunburst instead of Flare?

Flare prioritizes everyday generation and responsiveness. Consider Sunburst for demanding visual requirements or precise edits. Both models generate and edit images; their positioning reflects a quality and latency tradeoff.

Models & Capabilities
Which output options does Flare offer on HiAPI?

HiAPI offers six quality settings, square and landscape or portrait formats, and listed pixel dimensions. Choose PNG, JPEG or WebP, using PNG or WebP for transparency. The specifications and Playground list the options.

Models & Capabilities
How can Flare fit into a repeated-generation workflow?

An application can submit image tasks and retrieve results through polling or callbacks. Each HiAPI request produces one image, so the application manages task volume and concurrency.

Models & Capabilities
How is Flare billed?

Billing is per output image and selected quality, with no extra reference-image fee. Playground displays the estimate for the chosen settings.

Models & Capabilities
What is GPT Image 2.5 Sunburst best suited to?

Sunburst targets visual work with demanding quality and precision requirements, including campaigns, refined product imagery, multiple references and focused revisions. It also generates new images directly from text.

Models & Capabilities
What is the main tradeoff between Sunburst and Flare?

Sunburst emphasizes quality and precision, generally with longer generation times. Flare focuses on faster everyday creation. The choice depends on the visual requirements and acceptable wait.

Models & Capabilities
Can Sunburst combine a person with several clothing references?

Yes. The outfit example uses a person, jacket, top and boots as four inputs, combining the garments within the person’s original setting. HiAPI supports up to 16 references.

Models & Capabilities
Can it remove a single object from a photograph?

Yes. Specify the object to remove and describe the person and setting to preserve. The portrait example targets the flower in the subject’s hand, illustrating photo cleanup, prop changes and content revisions.

Models & Capabilities
Can Sunburst replace furniture in an interior image?

The kitchen example replaces chairs while retaining the table, windows and room context. Similar edits support furniture styling and interior proposals.

Models & Capabilities
Can Sunburst translate text inside an image?

The coffee-machine example shows an English diagram revised into Spanish. These edits support localized diagrams, international promotional assets and language updates to existing designs.

Models & Capabilities
How does an advertising concept carry through successive edits?

The product sequence begins with a bottle photo, creates a sunset billboard and then edits that output into a snowy winter scene. Each iteration uses the relevant image as input with the next requested change.

Models & Capabilities
Can Sunburst generate complex packaging from text?

It can create product concepts combining a package shell, printed wording, materials and decoration. The toy-airplane example illustrates this kind of brief for packaging presentations, product visuals and creative proposals.

Models & Capabilities
Can isolated products be used in transparent compositions?

Sunburst supports isolated subjects and transparent backgrounds. On HiAPI, choose a transparent background with PNG or WebP. Use the output in advertising compositions, product displays and page designs.

Models & Capabilities
What does the lacquer-cabinet example demonstrate?

The requested change targets the central lunar inlay on the cabinet door. The surrounding ornament, cabinet and tools form the context to preserve, illustrating editing around one defined local objective.

Models & Capabilities
Which quality and size options are available for Sunburst?

HiAPI offers low, medium, high, xhigh, max and auto, along with the listed aspect ratios and pixel dimensions. Quality and size are independent, and the default quality is medium.

Models & Capabilities
Is each Sunburst revision billed separately?

Each submitted revision is an independent image task billed by its selected quality and output image, without an extra reference-image fee. The page shows an estimate and provides preview and download after completion.

Models & Capabilities
What is GPT Image 2?

GPT Image 2 is an OpenAI text-to-image model for posters, knowledge cards, product visuals, and images with explicit text and layout requirements.

Models & Capabilities GPT Image 2
Is GPT Image 2 the same as GPT Image 2.0?

GPT Image 2.0 is a common search shorthand; use the exact GPT Image 2 model ID shown on this page and in the API docs.

Models & Capabilities GPT Image 2
Which endpoint and model ID does the GPT Image 2 API use?

Send POST requests to /v1/tasks, set model to gpt-image-2/text-to-image, and pass prompt, aspect_ratio, and resolution in input. The create response returns a taskId; retrieve the final image through task polling or a callback.

Models & Capabilities GPT Image 2
How do I use GPT Image 2?

In the Playground, describe the subject, scene, layout, and exact text, then choose the aspect ratio, resolution, and background. Click Run to view the result.

Models & Capabilities GPT Image 2
Which resolutions and aspect ratios does GPT Image 2 support?

It supports 1K, 2K, and 4K plus 16 aspect ratios including auto, 1:1, 3:2, 2:3, 4:3, 3:4, 16:9, and 9:16. The auto or omitted aspect ratio supports 1K only; 1:1 does not support 4K; 2K does not support 5:4, 4:5, 3:1, 1:3, or 9:21; 4K does not support 1:1, 3:1, 1:3, or 9:21. An explicitly provided background requires 1K.

Models & Capabilities GPT Image 2
What background options are available?

background accepts auto, opaque, or transparent; the background parameter is supported only at 1K, so omit it for 2K/4K.

Models & Capabilities GPT Image 2
How do I use GPT Image 2 with HiAPI?

Validate the prompt and composition in the HiAPI Playground, then call POST /v1/tasks to create an asynchronous text-to-image task and retrieve the result by taskId or callback.

Models & Capabilities GPT Image 2
Which languages and layout instructions does GPT Image 2 support?

Specify the target language, exact copy, typography style, placement, hierarchy, and overall layout in the prompt. Write the text you want rendered verbatim for complex copy.

Models & Capabilities GPT Image 2
How do I choose between GPT Image 2 text-to-image and image-to-image?

Use gpt-image-2/text-to-image when generating from a description without references. Use gpt-image-2/image-to-image when preserving or editing an existing visual with 1–6 reference images.

Models & Capabilities GPT Image 2
Does every clip include sound?

Yes. Sound is generated together with the picture and cannot be switched off. Describe the ambience, music or dialogue you want in the prompt.

Models & Capabilities Grok Imagine 1.5 Lite
Which durations, resolutions and aspect ratios can I use?

Any whole-second duration from 1 to 15 seconds (default 6), 480p, 720p or 1080p (default 720p), and 16:9, 4:3, 3:2, 1:1, 2:3, 3:4 or 9:16 (default 16:9).

Models & Capabilities Grok Imagine 1.5 Lite
How long can the prompt be?

Up to 4,096 characters. A focused paragraph about subject, motion, camera and sound usually works better than a long list.

Models & Capabilities Grok Imagine 1.5 Lite
Can I animate my own image with this model?

No, this route takes text only. Use grok-imagine-1.5-lite/image-to-video when the shot should start from your image.

Models & Capabilities Grok Imagine 1.5 Lite
How do I call it through the API?

POST an asynchronous task to /v1/tasks with model set to grok-imagine-1.5-lite/text-to-video and prompt, duration, resolution and aspect_ratio inside input. The request returns a task ID; poll the task or use a callback to get the video URL.

Models & Capabilities Grok Imagine 1.5 Lite
How is it billed?

By the requested duration at the selected resolution. The pricing panel shows the live rate, and the successful task record shows the final charge.

Models & Capabilities Grok Imagine 1.5 Lite
How many images can I upload?

Exactly one, passed as image_url. It must be publicly accessible.

Models & Capabilities Grok Imagine 1.5 Lite
Can I choose the output aspect ratio?

No. The video approximately follows the aspect ratio of your image, with dimensions rounded to sizes the model supports. Crop or extend the image to the ratio you need before submitting.

Models & Capabilities Grok Imagine 1.5 Lite
Which durations and resolutions are supported?

Any whole-second duration from 1 to 15 seconds (default 6) at 480p, 720p or 1080p (default 720p).

Models & Capabilities Grok Imagine 1.5 Lite
Is a prompt required?

Yes. This route requires both image_url and a prompt of up to 4,096 characters. Describe what moves, the camera and the sound rather than re-describing the image.

Models & Capabilities Grok Imagine 1.5 Lite
How do I keep the subject stable?

Use a clear image with a complete composition, describe one main motion, and name what should stay unchanged, such as faces, the art style or the background layout.

Models & Capabilities Grok Imagine 1.5 Lite
What is Grok Imagine Quality best for?

Hero images, covers, campaign concepts, and high-quality scenes that need richer material, light, atmosphere, and detail.

Models & Capabilities Grok Imagine Quality
How are 1k and 2k billed?

1k and 2k use separate per-image tiers. Ratio and output format do not add another fee; see the live page price for the current amounts.

Models & Capabilities Grok Imagine Quality
Why not always use the standard tier?

Standard is better for volume drafts and speed. Choose Quality when final-image material and detail matter more.

Models & Capabilities Grok Imagine Quality
Does it accept reference images?

The text-to-image route does not. Use grok-imagine-quality/image-to-image for quality-tier editing.

Models & Capabilities Grok Imagine Quality
How do I compare resolutions fairly?

Keep prompt, ratio, and format fixed, change only resolution, and save both 1k and 2k outputs.

Models & Capabilities Grok Imagine Quality
What is the standard Grok Imagine route best for?

It suits high-volume ideation, social assets, landscapes, and cinematic scenes. Current 1k and 2k output share one flat price.

Models & Capabilities Grok Imagine
How does standard differ from Quality?

Standard emphasizes speed and a low flat price. Quality is a separate model route focused on material, detail, and light, with tiered 1k/2k pricing.

Models & Capabilities Grok Imagine
Which ratios are available?

Thirteen landscape, square, and portrait ratios ranging from 2:1 and 20:9 to 1:2.

Models & Capabilities Grok Imagine
Does this page accept references?

No. Use grok-imagine/image-to-image for reference editing.

Models & Capabilities Grok Imagine
What is required in the API request?

Submit model and input.prompt. Ratio, resolution, and format may use defaults.

Models & Capabilities Grok Imagine
What inputs does HeyGen Avatar IV need?

One portrait photo plus one driving method: a script with a chosen voice, or an audio file. With a script and no voice selected, the default voice Warm Pro Narrator is used.

Models & Capabilities HeyGen Avatar IV
What kind of photo works best?

A single, front-facing person with a clear, unobstructed face, even lighting and a relaxed closed mouth gives the steadiest result. Profiles, masks, group photos and tiny faces do not work well.

Models & Capabilities HeyGen Avatar IV
How long is the output video?

No duration parameter is needed. With a script the length follows the text and speaking pace; with audio it matches the uploaded file, which can be up to 600 seconds.

Models & Capabilities HeyGen Avatar IV
Which aspect ratios and resolutions are supported?

Aspect ratios 16:9, 9:16, 4:5, 5:4, 1:1 and auto (keeps the photo's ratio); resolutions 720p and 1080p.

Models & Capabilities HeyGen Avatar IV
When do I need a background?

Only when the output ratio differs from the photo and the frame has empty space. Choose a #RRGGBB color or a background image; white is used otherwise.

Models & Capabilities HeyGen Avatar IV
How is it different from HeyGen Avatar V?

Avatar IV animates the photo you upload; Avatar V uses one of 497 preset avatars and additionally supports transparent WebM, background removal and SRT captions.

Models & Capabilities HeyGen Avatar IV
How is HeyGen Avatar IV billed?

By the real seconds of generated video. A hold based on the estimated length is taken at creation and settled to the real length afterwards, with failed tasks refunded in full; see the live price on this page.

Models & Capabilities HeyGen Avatar IV
What are the mask requirements?

The mask must match the source aspect ratio. Black areas are repainted and white areas are kept; a mask slightly larger than the silhouette looks more natural.

Models & Capabilities Ideogram V3 Character
What size is the output?

The output keeps the source size; a 1024×768 source returns 1024×768.

Models & Capabilities Ideogram V3 Character
Can I replace a person already in the image?

Yes. Paint that person into the mask and describe the new character's action; people and background outside the mask are kept.

Models & Capabilities Ideogram V3 Character
What format should the mask be?

An image with the same aspect ratio as the source, with the area to repaint in black on white; the examples on this page use PNG masks.

Models & Capabilities Ideogram V3 Character
How large should the mask be?

Slightly larger than the character silhouette. Too small cramps the pose; too large repaints more of the background.

Models & Capabilities Ideogram V3 Character
What should the prompt say?

Only what the character does inside the mask, such as sitting on the bench with an umbrella or waving by the window. There is no need to describe the background; everything outside the mask stays.

Models & Capabilities Ideogram V3 Character
Can I place the character in different spots of one image?

Yes. Keep the source and reference, and change the mask position and the action in the prompt; the cabin armchair and window versions on this page were made that way.

Models & Capabilities Ideogram V3 Character
Can I skip the mask?

No. image_url, mask_url and reference_image_urls are all required; without a source image, use the Character generation model.

Models & Capabilities Ideogram V3 Character
Can I send more than one character reference?

No. reference_image_urls must contain exactly one image; illustrations, mascots and photos all work.

Models & Capabilities Ideogram V3 Character
Do rendering_speed and style work for edits?

Yes. The TURBO / BALANCED / QUALITY tiers and AUTO / REALISTIC / FICTION styles apply as well; each tier is priced per image, see live pricing.

Models & Capabilities Ideogram V3 Character
Is Remix supported?

Not yet. Character generation (ideogram-v3-character) and masked editing (ideogram-v3-character/edit) are available today.

Models & Capabilities Ideogram V3 Character
Am I charged for failures, and how do I get results?

Failed tasks are not charged. Save the taskId, query GET /v1/tasks/:id and download the WebP from output[].url when status is success; output[].url is temporary, so save results promptly.

Models & Capabilities Ideogram V3 Character
Can I choose the output aspect ratio?

Not separately. The output follows the character reference: portrait references return 864×1152 and landscape references return 1328×752. Upload a landscape reference (for example by padding the background) for landscape output.

Models & Capabilities Ideogram V3 Character
What makes a good reference?

A clear frontal or three-quarter view on a simple background; illustrations and photos both work. Outfits and colors in the reference are kept where possible.

Models & Capabilities Ideogram V3 Character
How should I write the prompt?

Refer to the subject with a fixed phrase (the explorer character / the same man) and focus on scene, action, props and light; pair it with style REALISTIC or FICTION.

Models & Capabilities Ideogram V3 Character
Can the character be an animal or brand mascot?

Yes. The orange fox mascot on this page comes from one character design and appears on a billboard, behind a counter and at a fair with the same scarf and colors.

Models & Capabilities Ideogram V3 Character
Can I build a sequential story or storyboard?

Yes. Use the same character reference for every frame and change only the scene and action in the prompt, as in the four-frame story on this page.

Models & Capabilities Ideogram V3 Character
How do the three style options differ?

AUTO lets the model decide from the prompt; REALISTIC leans toward photographic light and texture for portraits and real scenes; FICTION keeps an illustrated look for picture books, games and cartoon characters.

Models & Capabilities Ideogram V3 Character
What is seed for?

A fixed seed makes results reproducible and comparable: keep the reference, prompt and seed, then switch only the tier or style to compare.

Models & Capabilities Ideogram V3 Character
How long are result images kept?

output[].url is a temporary link with an expiry; download or copy results to your own storage after the task succeeds.

Models & Capabilities Ideogram V3 Character
Which rendering_speed should I choose?

TURBO is fastest and lowest-priced for exploring directions and drafts; BALANCED is the default; QUALITY gives the finest detail for final images. All three use the same model ID and are priced per image; see live pricing.

Models & Capabilities Ideogram V3 Character
What edits can Ideogram 4.5 Edit do?

Four kinds: prompt edits, reference restyles, mask inpainting and Precise Edit. They can be combined.

Models & Capabilities Ideogram V4.5
Which fields are required?

prompt and image_url. Add reference_image_urls, mask_url, edit_precision, quality, image_size, num_images and seed as needed.

Models & Capabilities Ideogram V4.5
How do regular and Precise Edit differ?

Use Precise Edit for text changes and small fixes, and regular for recolors and restyles. edit_precision=high is Precise Edit and restores unedited pixels; regular is freer.

Models & Capabilities Ideogram V4.5
Do both precisions share quality tiers?

No. regular offers very_low, low, medium and high; Precise Edit offers low, medium and high.

Models & Capabilities Ideogram V4.5
How do I make a mask?

Make it the same size as the source, with black for the area to repaint and white for what to keep. Make the black area slightly larger than the object.

Models & Capabilities Ideogram V4.5
When must image_size be auto?

With mask_url or Precise Edit. The output then follows the source size; only regular edits without a mask accept the six presets.

Models & Capabilities Ideogram V4.5
How many reference images can I send?

Up to 4, or 3 when a mask is provided. Say what each reference contributes in the prompt.

Models & Capabilities Ideogram V4.5
How is it billed?

Per output image. The rate depends on quality and edit_precision, and each request returns one image; size, references and masks do not change the price. See the live price for current rates.

Models & Capabilities Ideogram V4.5
How do I write an edit instruction?

Use four parts: the task, what to change, what to keep, and the finish and limits. For example: ① Edit the provided logo. ② Change the text "Ember & Oak" to "Maple & Co". ③ Keep the serif lettering, bean emblem, tagline, colors and layout. ④ The result should look like the original print; do not add new elements. Precise Edit works best for text changes.

Models & Capabilities Ideogram V4.5
Is this the same as Ideogram 4.5 Image to Image?

Yes. Only the display name changed to Ideogram 4.5 Edit; the model ID ideogram-v4.5/image-to-image and page link are unchanged.

Models & Capabilities Ideogram V4.5
What if I have no source image?

Generate one with Ideogram 4.5 (ideogram-v4.5/text-to-image) first. Then edit it here.

Models & Capabilities Ideogram V4.5
How were the examples made?

HiAPI generated them all with this model. Precise Edit used low and the rest used regular very_low, with source images from the Ideogram 4.5 page.

Models & Capabilities Ideogram V4.5
What is Ideogram 4.5 good for?

Images that need clear text. Posters, logos and wordmarks, packaging labels, signage and magazine covers are its strengths.

Models & Capabilities Ideogram V4.5
How do I get more accurate text?

Put the exact words in quotes. Then say whether each line is a headline, subtitle or small print, plus placement and type feel; shorter text is more reliable.

Models & Capabilities Ideogram V4.5
How should I write a prompt?

Use four parts: the task, the text and details, the layout to keep, and the finish and limits. For example: ① Create a conference poster. ② Headline "DESIGN FORWARD", subtitle "Typography Summit 2026", orange and navy palette. ③ Keep the headline centered and the subtitle clearly smaller. ④ Crisp lettering, flat print look; do not add any other text.

Models & Capabilities Ideogram V4.5
How should I choose a quality tier?

Use medium for everyday work, low for exploring and high for finals. Each tier has its own price, and high has the most detail for fine print.

Models & Capabilities Ideogram V4.5
Which sizes are supported?

Six presets: square, square_hd (default), portrait_4_3, landscape_4_3, portrait_16_9 and landscape_16_9. Custom sizes are not offered here.

Models & Capabilities Ideogram V4.5
Does size change the price?

No. Billing is per image and only quality changes the rate; each request returns one image.

Models & Capabilities Ideogram V4.5
What does enable_prompt_expansion do?

It lets the model expand your prompt before drawing, and it is on by default. Keep it on for short prompts; turn it off when the layout must follow your text strictly.

Models & Capabilities Ideogram V4.5
How many images per request?

One. num_images is fixed at 1; to compare several layouts, send a few requests with different seeds.

Models & Capabilities Ideogram V4.5
How do I reproduce an image?

Pass a fixed seed. Keep the prompt, size and quality unchanged as well.

Models & Capabilities Ideogram V4.5
How well does it handle Chinese text?

Large Chinese headlines usually come out right, but small Chinese text can slip at low. Use medium or high when it matters.

Models & Capabilities Ideogram V4.5
Can I upload a reference image?

No, this model takes text only. To modify an existing image, use Ideogram 4.5 Edit (ideogram-v4.5/image-to-image).

Models & Capabilities Ideogram V4.5
How do I keep the output?

Download it from the result image URL. URLs expire, so copy results to your own storage promptly.

Models & Capabilities Ideogram V4.5
What is Imagen 4 good for?

Everyday product, lifestyle and content images. Describe the subject, scene, material and light to get an image ready for review, for product heroes, campaign drafts and editorial illustrations.

Models & Capabilities Imagen 4
Which aspect ratios does Imagen 4 support?

Six: 1:1, 16:9, 9:16, 3:4, 4:3 and auto. aspect_ratio is required; the Playground on this page defaults to 16:9.

Models & Capabilities Imagen 4
How long can the prompt be?

Up to 5,000 characters. prompt is required; order it as subject, scene, material, light and text.

Models & Capabilities Imagen 4
Can I exclude things I don't want?

Yes, with negative_prompt. It is optional, up to 5,000 characters, and works for extra text, watermarks or a cluttered background.

Models & Capabilities Imagen 4
What format does seed use?

A string of up to 500 characters, optional. Keep the seed and other inputs fixed to compare only the prompt change.

Models & Capabilities Imagen 4
Can I upload a reference image?

No. Imagen 4 generates from text only; to edit an existing image, see Nano Banana 2.1 in the related models, which edits with references.

Models & Capabilities Imagen 4
How many images per request?

One. Submit several tasks for more versions.

Models & Capabilities Imagen 4
Can I set pixel size or output format?

No. Only aspect_ratio is exposed; custom pixel dimensions and output-format controls are not offered.

Models & Capabilities Imagen 4
Can the image contain text?

Yes, but check it before delivery. Put the exact text in quotes with its position and size and say no other text should appear; the poster example on this page came back with one extra letter in its headline.

Models & Capabilities Imagen 4
Imagen 4, Fast or Ultra?

Choose Imagen 4 for everyday production. Choose Imagen 4 Fast for many drafts and Imagen 4 Ultra when the final needs more detail; each has its own model ID, so one project can use them at different stages.

Models & Capabilities Imagen 4
How do I call the API?

Submit a task to POST /v1/tasks. Send model: "imagen-4.0" and prompt plus aspect_ratio inside input. The response returns a taskId right away; set callback.url for a completion notice or query GET /v1/tasks/:id, then download the image from output.

Models & Capabilities Imagen 4
How is it billed?

Per image, one per request. Aspect ratio and seed do not change billing; see the live price on this page, and the task record is final.

Models & Capabilities Imagen 4
What is Imagen 4 Fast good for?

Work that needs many versions. Use it to pick creative directions, run batch drafts, and compare color and composition for social or e-commerce.

Models & Capabilities
Does Fast take the same inputs as Imagen 4?

Almost; only the seed type differs. Both use prompt, aspect_ratio, negative_prompt and seed; seed is an integer on Fast and a string on Imagen 4 and Ultra.

Models & Capabilities
Which aspect ratios does Fast support?

Six: 1:1, 16:9, 9:16, 3:4, 4:3 and auto. aspect_ratio is required; the Playground on this page defaults to 16:9.

Models & Capabilities
How do I use seed?

Pass an integer, optional. In batch drafts, fix the seed and change one thing in the prompt to see exactly what that change does.

Models & Capabilities
How long can the prompt be?

Up to 5,000 characters. When drafting, start with a short, clear subject and scene, then add detail round by round.

Models & Capabilities
Can I exclude things I don't want?

Yes, with negative_prompt. It takes up to 5,000 characters and keeps extra text or watermarks out of a whole batch.

Models & Capabilities
Can I upload a reference image?

No. Imagen 4 Fast generates from text only; for references, see Nano Banana 2 Lite in the related models, which accepts several.

Models & Capabilities
How many images per request?

One. For batches, submit tasks in parallel and receive results with callback.url.

Models & Capabilities
Can I set pixel size or output format?

No. Only aspect_ratio is exposed; custom pixel dimensions and output-format controls are not offered.

Models & Capabilities
How do I make the final once a direction is chosen?

Just change model. Send the same prompt and ratio to Imagen 4 or Imagen 4 Ultra with that model ID. Note that seed is a string on those two models.

Models & Capabilities
How do I call the API?

Submit a task to POST /v1/tasks. Send model: "imagen-4.0-fast" and prompt plus aspect_ratio inside input. The response returns a taskId right away; set callback.url or query GET /v1/tasks/:id, then download the image from output.

Models & Capabilities
How is it billed?

Per image, one per request. See the live price on this page; the Playground shows an estimate before you submit.

Models & Capabilities
What is Imagen 4 Ultra good for?

Final delivery. It suits product macros, portraits, landscapes and material close-ups, and images that need to follow the prompt closely.

Models & Capabilities
Which aspect ratios does Ultra support?

Six: 1:1, 16:9, 9:16, 3:4, 4:3 and auto. aspect_ratio is required; the Playground on this page defaults to 1:1.

Models & Capabilities
How long can the prompt be?

Up to 5,000 characters. At the final stage, spell out material, light direction, depth of field and background.

Models & Capabilities
Can I exclude things I don't want?

Yes, with negative_prompt. It takes up to 5,000 characters; use it for blown highlights, extra text, watermarks or a cluttered background.

Models & Capabilities
What format does seed use?

A string of up to 500 characters, optional. Fix it while fine-tuning the prompt to compare detail changes.

Models & Capabilities
Can I upload a reference image?

No. Imagen 4 Ultra generates from text only; to start from an existing image, see Nano Banana Pro in the related models, which accepts references.

Models & Capabilities
How many images per request?

One. Submit several tasks for more candidates.

Models & Capabilities
Is text in the image reliable?

Not guaranteed, so zoom in before delivery. Write each exact phrase with its position and size and say no other text should appear for steadier results.

Models & Capabilities
When is Ultra worth it?

When the direction is settled and you need a finished image. Explore with Imagen 4 Fast first, then send the chosen prompt and ratio to Ultra by changing model to imagen-4.0-ultra.

Models & Capabilities
How do I call the API?

Submit a task to POST /v1/tasks. Send model: "imagen-4.0-ultra" and prompt plus aspect_ratio inside input. The response returns a taskId right away; set callback.url or query GET /v1/tasks/:id, then download the image from output.

Models & Capabilities
How is it billed?

Per image, one per request. See the live price on this page; the task record is final.

Models & Capabilities
Which Kimi K3 reasoning level should I choose?

Start with low for extraction, rewriting, and short questions; try high for analysis with several steps, and compare max on difficult reasoning or coding tasks. reasoning_effort defaults to max and offers only low, high, and max, without medium or none. Higher effort may generate more reasoning tokens and take longer, so compare answer quality and usage on your task.

Models & Capabilities Kimi K3
Why are there no thinking-off or temperature controls?

Thinking is always enabled for this Kimi K3 integration. Try low to reduce work instead of sending thinking.type=disabled. The Playground omits thinking, temperature, and top_p. When migrating from another model, remove those controls and select an effort using the top-level reasoning_effort field.

Models & Capabilities Kimi K3
Why can a short answer still incur substantial output-token usage?

Generated reasoning is also billed as output, so the visible answer length is not the total output count. usage.completion_tokens already includes reasoning tokens; completion_tokens_details.reasoning_tokens, when returned, is a breakdown rather than an extra charge to add. Calculate ordinary input, cache reads, and output separately: divide each token count by one million and multiply by its current USD rate shown on this page.

Models & Capabilities Kimi K3
Can I keep only the final answer when continuing a conversation?

Keep the required conversation history in order and replay the complete assistant message returned by the API, including content, any reasoning_content, and any tool_calls. Saving only the final text discards context used for reasoning or tool continuation. This Playground preserves the reasoning content returned by the model.

Models & Capabilities Kimi K3
Are tools executed automatically when Kimi returns tool_calls?

No. The model proposes a function and its arguments; your application executes it. Append the original assistant message and a role=tool result message to messages. Its tool_call_id must match the original call id, then request the next response. Do not send the tool result only as an ordinary user message.

Models & Capabilities Kimi K3
How do JSON mode and strict JSON Schema differ?

response_format.type=json_object requests JSON, while your application still validates its business fields. To constrain fields and types, use response_format.type=json_schema with response_format.json_schema.schema and response_format.json_schema.strict=true. Parse the final JSON from choices[0].message.content, without concatenating reasoning_content. The JSON output switch in this Playground uses json_object.

Models & Capabilities Kimi K3
Can I send images or video because the official Kimi model supports vision?

This HiAPI Kimi K3 integration currently exposes text input and text output only. The official model's visual capabilities do not mean image or video input is enabled here. Use text messages rather than copying image_url or video_url examples from the provider. Any future input support will be reflected in this page's capability list and Playground controls.

Models & Capabilities Kimi K3
Can I use the Responses API or migrate by changing only the model name?

This integration currently exposes POST /v1/chat/completions with model=kimi-k3; Responses is not an enabled endpoint for it. With the OpenAI SDK, use chat.completions.create and messages, update the reasoning fields, and preserve complete assistant history. You can reuse one HiAPI API key, but existing Responses code also needs its request and response handling adapted.

Models & Capabilities Kimi K3
Why does repeating the same context not always produce the same cache charge?

Cache-read billing uses the actual usage.prompt_tokens_details.cached_tokens returned for the request. Repeated text does not guarantee a hit. Keep the initial system prompt, tool definitions, and fixed context stable, then append new messages; changing reasoning_effort can also affect cache hits. When cached_tokens is zero, do not estimate that input at the cache-read rate.

Models & Capabilities Kimi K3
Does a 1M context window mean a single response can contain one million tokens?

No. Context size and the maximum output for one request are different limits. This page does not infer an output limit from the 1M context window and displays “-” for an unconfirmed maximum. Set max_tokens as the output budget for a request. If finish_reason is length, check whether an output limit was reached before splitting the task or continuing the response.

Models & Capabilities Kimi K3
How should I prepare the first-frame image?

Use jpeg, png or webp up to 10MB; jpeg and png are the safest choices. Try to match the image to the ratio you plan to output. When they differ, the clip may be center-cropped or may keep the image's own proportions, so keep the subject near the middle.

Models & Capabilities
What does the last frame do, and how do I use it?

The last frame is an optional second image that decides where the clip ends. Keep the background and camera position the same in both images and change only what should change, like the red car turning blue on this page. With a last frame, choose 16:9, 9:16 or 1:1.

Models & Capabilities
Which lengths and aspect ratios can I choose?

Pick 5 or 10 seconds. Ratios are 16:9, 9:16, 1:1, 4:3, 3:4 and 21:9, rendering at 1920×1080, 1080×1920, 1440×1440, 1660×1244, 1244×1660 and 2200×940. Resolution follows the ratio, so there is nothing else to set.

Models & Capabilities
Do the videos have sound?

No. Clips come without an audio track, so add music or voice-over in your edit. If you want sound built in, Kling 2.5 Turbo Text-to-Video generates from text and every clip includes audio.

Models & Capabilities
What should the prompt describe?

The first frame already shows the subject and setting, so describe motion, changes in the scene and the camera, for example “petals drift down, slow orbit”. You can also list things you don't want to see and adjust how closely the video follows the prompt.

Models & Capabilities
How long are generated videos kept?

Output links last about 7 days by default. Download anything you want to keep, or choose long-term storage when you create the task; past runs stay in your task history.

Models & Capabilities
Kling 2.5 Turbo Image-to-Video or Text-to-Video: which should I use?

Use Image-to-Video when you have a picture and want the subject and composition to follow it. Use Text-to-Video when you only have an idea, or when the clip needs sound. Both run 5 or 10 seconds.

Models & Capabilities
Do I need to upload an image?

No. This version takes text only, up to 2,500 characters; put the subject, setting, action and camera in one passage. If the picture should follow an image you already have, use Kling 2.5 Turbo Image-to-Video.

Models & Capabilities
Do the clips have sound, and can I turn it off?

Every clip includes an audio track, and there is currently no way to switch it off. The sound is generated by the model and may not match your brief, so listen first and replace or lower it in your edit.

Models & Capabilities
Which lengths and aspect ratios are available?

5 or 10 seconds, in 16:9, 9:16 or 1:1, rendering at 1920×1080, 1080×1920 and 1440×1440. Resolution follows the ratio, so there is no separate setting.

Models & Capabilities
How do I write a prompt that works?

Write one passage in the order subject, setting, action, camera, then add the light. One or two actions are plenty, like the raised paw and the turn in the red panda clip; ask for too much and five seconds won't fit it all.

Models & Capabilities
How do I keep unwanted things out of the video?

List them separately as things you don't want to see, such as text, logos or extra people. You can also adjust how closely the video sticks to your description.

Models & Capabilities
Kling 2.5 Turbo or Kling 3.0 Turbo Text-to-Video: which should I use?

Use Kling 2.5 Turbo when 5 or 10 seconds is enough. Use Kling 3.0 Turbo Text-to-Video for any length from 3 to 15 seconds or a choice of 720p and 1080p. Both offer landscape, portrait and square.

Models & Capabilities
What inputs does Kling 3.0 Motion Control need?

Two required inputs: a character reference image (image_url) and a 3-30 second motion reference video (video_url). The image defines appearance and identity; the clip defines motion, expressions and timing. A prompt is optional for scene, style and camera guidance.

Models & Capabilities
How is the output length decided?

No duration parameter is needed. Output follows the reference clip: up to 30s with character_orientation=video and up to 10s with image. Very fast or complex motion may yield a shorter clip because only usable motion segments are extracted.

Models & Capabilities
When should I use video vs image orientation?

video: the character faces the way the reference clip does, best for dance, walking and multi-angle motion. image: the character keeps the orientation of the reference image, best for fixed-camera talking or explainer content, capped at 10s.

Models & Capabilities
What does the reference video need?

mp4/mov, up to 100MB, 3-30 seconds, one continuous shot with the subject's head, shoulders and torso clearly visible; prefer a single person, no cuts and no occlusion. The image also needs head, shoulders and torso visible, both sides over 340px, aspect ratio between 2:5 and 5:2.

Models & Capabilities
Is the reference audio kept?

Yes by default (keep_original_sound=true), which suits talking and dance covers that reuse the original track; pass false for video-only output.

Models & Capabilities
How is Kling 3.0 Motion Control billed?

Per actual output second, with different rates for 720p and 1080p. A hold sized by the orientation cap is taken at creation and adjusted to the real output seconds after completion; see the live pricing block on this page.

Models & Capabilities
How long and how large can the video be?

Duration is 3–15 seconds in whole seconds (5 by default), resolution is 720p, 1080p or 4K, and the aspect ratio is 16:9, 9:16 or 1:1.

Models & Capabilities
Does the video have sound?

Turn on native audio (audio) and the model generates matching sound with the picture, such as effects, ambience and rhythm. It is off by default, and the output has no audio track when it is off.

Models & Capabilities
What is smart multi-shot?

With prefer_multi_shots on, the model plans several shots from the story in your prompt and cuts between them, which suits a complete short story. Leave it off when you want one continuous shot.

Models & Capabilities
How do I keep a fixed character or product in the video?

Add a subject to elements with a name, a short description and 2–4 images of the same subject, then mention it in the prompt as @name. Text-to-video accepts up to 7 subjects.

Models & Capabilities
How is it different from Kling 3.0 Text-to-Video?

Both cover 3–15 seconds and 720p to 4K. The Omni version adds subject references through elements and smart multi-shot, which helps when a character or product has to stay consistent. For plain text-only clips either one works.

Models & Capabilities
How is it billed?

Billing follows video duration; resolution and native audio change the rate. See the pricing page for current prices and your HiAPI task history for per-task charges.

Models & Capabilities
How many images do I upload?

Send 1–2 images in image_urls: one is the first frame; with two, order them as [first frame, last frame] so the second sets the ending. JPG or PNG, up to 50MB each, at least 300px, with an aspect ratio between 0.4 and 2.5.

Models & Capabilities
Can I choose the aspect ratio?

Not separately. The output frame follows the first image, so prepare a portrait or square first frame when you want that shape.

Models & Capabilities
How long can the video be?

Duration is 3–15 seconds in whole seconds (5 by default), at 720p, 1080p or 4K.

Models & Capabilities
Are sound and smart multi-shot supported?

Yes. Native audio generates matching sound with the picture, and smart multi-shot lets the model plan several shots from your prompt. Both are off by default.

Models & Capabilities
Can I reference subjects as well?

Yes. Add up to 3 subjects in elements, each with 2–4 images, and mention them as @name so a character or product stays consistent beyond the first frame.

Models & Capabilities
How many reference images can I upload?

Send 1–7 reference images in image_urls, or up to 4 with a reference video. Reference images and elements subjects together may not exceed 7 (4 with a reference video). JPG or PNG, up to 50MB each, at least 300px.

Models & Capabilities
What does the reference video do, and what are its limits?

video_urls takes one optional reference clip that guides camera motion and pacing. It must be MP4 or MOV, up to 200MB, 3–15.5 seconds, 700–4553px per side and 24–60fps. With a reference video, native audio and smart multi-shot are not available.

Models & Capabilities
How is reference-to-video different from image-to-video?

Image-to-video uses your image as the first frame. Reference-to-video uses images to define what a person, pet or object looks like, then builds a new composition and scene; the references do not appear as the opening frame.

Models & Capabilities
How long and in what frame?

3–15 seconds (5 by default), at 720p, 1080p or 4K, in 16:9, 9:16 or 1:1.

Models & Capabilities
How do I tell the model who is who?

Describe the subject in words, for example “the woman in the yellow raincoat from the references”. For a firmer binding, name the subject in elements and mention it as @name.

Models & Capabilities
What kind of video do I upload?

Send one source video in video_urls: MP4 or MOV, up to 200MB, 3–15.5 seconds, 700–4553px per side and 24–60fps.

Models & Capabilities
Can I set the duration or aspect ratio?

There is nothing to set. Output length and frame follow the source video; the option you choose is resolution (720p, 1080p or 4K).

Models & Capabilities
What is kept and what changes?

The source's motion and camera movement are kept, while style, setting, weather or subject change according to the prompt. Saying “keep the original motion and camera movement” in the prompt makes the result steadier.

Models & Capabilities
Does the output have sound?

No. Transformation does not generate audio; add sound in editing, or bring the source audio back in post.

Models & Capabilities
How do I swap in a specific character?

Add a subject in elements with a name, description and 2–4 images, then say in the prompt that @name replaces the original subject. Up to 7 subjects are accepted.

Models & Capabilities
Does Kling 3.0 Turbo text-to-video require an input image?

No. This page uses the pure text-to-video model kling-3.0-turbo/text-to-video and accepts a text prompt without an image. Use the separate Kling 3.0 Turbo Image-to-Video page when a clip must start from a source frame.

Models & Capabilities Kling 3.0 Turbo
Which durations, resolutions, and aspect ratios does Kling 3.0 Turbo support?

The current HiAPI contract supports 3–15 seconds at 720p or 1080p in 16:9, 9:16, or 1:1. The page cases cover 720p, five- and twelve-second clips, landscape, and vertical output. The 1080p and 1:1 options are supported contract values with no page example yet.

Models & Capabilities Kling 3.0 Turbo
How is Kling 3.0 Turbo text-to-video priced?

Billing depends on generated duration and resolution, with separate per-second tiers for 720p and 1080p. Check the live page price and task record for the final charge.

Models & Capabilities Kling 3.0 Turbo
How should I write a Kling 3.0 Turbo text-to-video prompt?

Start with subject and setting, describe action and camera movement in order, then add lighting, style, and stability constraints. For a long take, specify single continuous shot, no cuts, and the camera path. For vertical work, mention vertical composition and set aspect_ratio to 9:16.

Models & Capabilities Kling 3.0 Turbo
Can I control audio in Kling 3.0 Turbo?

The current public request schema has no separate sound switch or audio field, so this page does not promise precise control of dialogue, music, or foley. A result file may contain an audio track, but production work should inspect the output and add voice, mix, and rights review in post when needed.

Models & Capabilities Kling 3.0 Turbo
How do I call the Kling 3.0 Turbo text-to-video API?

Send an asynchronous task to POST /v1/tasks, set model to kling-3.0-turbo/text-to-video, and include prompt, duration, resolution, and aspect_ratio in input. The create request returns a taskId. See the API docs for retrieval, callbacks, errors, and the complete parameter contract.

Models & Capabilities Kling 3.0 Turbo
What does the first-frame image need to be?

A publicly reachable jpeg or png URL inside the image_urls array, one image per request. That image becomes the opening frame of the clip.

Models & Capabilities Kling 4.0
What happens if I leave aspect ratio out?

Output follows the source image's ratio, so a portrait image stays portrait. Set aspect_ratio only when you want to override the original framing.

Models & Capabilities Kling 4.0
Is the prompt required?

Yes. The first frame supplies composition and identity; the prompt directs motion, camera and sound. Both are required, and shorter prompts usually work better.

Models & Capabilities Kling 4.0
How is image-to-video billed?

Per output second, same as text-to-video, with separate tiers for 720p and 1080p and no minimum charge.

Models & Capabilities Kling 4.0
Can the clip include audio?

Yes. Audio is generated together with the picture and the MP4 carries a soundtrack. Describe the sound you want in the prompt.

Models & Capabilities Kling 4.0
How do I call it through the API?

Send an asynchronous task to POST /v1/tasks with model set to kling-4.0/image-to-video and image_urls plus optional prompt, duration, resolution and aspect_ratio in input.

Models & Capabilities Kling 4.0
Does Kling 4.0 text-to-video need an input image?

No. kling-4.0/text-to-video takes a text prompt only. Use kling-4.0/image-to-video when a specific first frame has to come alive.

Models & Capabilities Kling 4.0
Which durations, resolutions and framings are supported?

Any whole duration from 3 to 30 seconds at 720p or 1080p, framed at 16:9, 9:16, 1:1 or 21:9 ultrawide. The examples here show five-second landscape clips at 720p and 1080p.

Models & Capabilities Kling 4.0
Does Kling 4.0 generate audio?

Yes. Picture and audio are produced in the same pass and the MP4 ships with a soundtrack. Ambient sound and sound sources written into the prompt, such as rain, an engine or an instrument, are used by the model.

Models & Capabilities Kling 4.0
How is text-to-video billed?

Per output second, with separate per-second tiers for 720p and 1080p. There is no minimum charge: a five-second clip costs five seconds at the tier rate. The live page price and task record show the exact amount.

Models & Capabilities Kling 4.0
How should I write a Kling 4.0 prompt?

Start with subject and setting, state the camera move explicitly (push-in, lateral tracking, handheld, locked-off), then add lighting and mood. Put the sound you want in the same description instead of planning to add it later.

Models & Capabilities Kling 4.0
What is Krea 2 Medium Turbo good for?

It turns a text description into a high-fidelity image very quickly, which suits work that needs fast iteration on prompts, such as concept sketches, social visuals and cinematic frames.

Models & Capabilities Krea 2 Medium Turbo
Which aspect ratios are supported?

aspect_ratio accepts eight values: 1:1 (default), 4:3, 3:2, 16:9, 2.35:1, 4:5, 2:3 and 9:16.

Models & Capabilities Krea 2 Medium Turbo
What is the 2.35:1 ratio for?

2.35:1 is a common widescreen cinema ratio, suited to cinematic scenes, website banners and video covers that need an extra-wide frame.

Models & Capabilities Krea 2 Medium Turbo
What does creativity do?

creativity controls how loosely the prompt is interpreted: raw, low, medium or high. Lower values stay closer to the prompt; higher values are more creative and may drift from it.

Models & Capabilities Krea 2 Medium Turbo
What is the default creativity?

On HiAPI it defaults to low, which keeps results close to the prompt. Switch to medium or high for more variation.

Models & Capabilities Krea 2 Medium Turbo
How do I use seed?

seed is an optional integer. Leave it out for a random result each time; keep the seed and other inputs fixed to reproduce a result and compare prompt or creativity changes.

Models & Capabilities Krea 2 Medium Turbo
How many images per request?

One image per request; submit several tasks for more.

Models & Capabilities Krea 2 Medium Turbo
Is there a prompt length limit?

prompt is required and must be 1–5000 characters.

Models & Capabilities Krea 2 Medium Turbo
Do I need a reference image?

No. This is a pure text-to-image model driven by the prompt alone. Style reference images, preset styles and moodboards are not offered at this time.

Models & Capabilities Krea 2 Medium Turbo
How is it billed?

Billing is per output image, one image per request. See the live price on this page for the current rate.

Models & Capabilities Krea 2 Medium Turbo
How do I get the result?

POST /v1/tasks returns a taskId right away. In production set callback.url to receive the completion notice; while debugging poll GET /v1/tasks/:id, then download the image URL from output once the task succeeds.

Models & Capabilities Krea 2 Medium Turbo
Is there content safety screening?

Yes. Safety screening is on for every generation, and requests that break content rules may fail.

Models & Capabilities Krea 2 Medium Turbo
What is Krea 3 good for?

Photoreal shots, portraits, product images, food, architecture, illustration and concept art, plus posters with headline text. Even a short prompt yields a complete scene, which makes it handy for exploring directions.

Models & Capabilities Krea 3
What is the difference between berry and jam?

They are two variants of the same model that share most inputs (negative_prompt applies to berry only): Krea positions berry for quality and jam for speed, at the same price. Switch with input.variant; the default is berry. In side-by-side runs, the same seed gives different framing and color at similar quality, so it is worth trying both.

Models & Capabilities Krea 3
What is reasoning_level and which should I use?

It sets how much effort goes into understanding and expanding the prompt before drawing: low or medium (default). Use medium for short or abstract prompts and low when your prompt is already specific and you want speed.

Models & Capabilities Krea 3
Which sizes are supported?

image_size accepts six presets: square_hd (1024 × 1024, default), square (512 × 512), portrait_4_3 (768 × 1024), portrait_16_9 (576 × 1024), landscape_4_3 (1024 × 768) and landscape_16_9 (1024 × 576).

Models & Capabilities Krea 3
Can I get 2K or a custom width and height?

No. This page offers the 1K tier; auto and custom sizes are not supported and are rejected as invalid parameters without any charge.

Models & Capabilities Krea 3
How is it billed?

Per successfully generated image. All six sizes cost the same and berry and jam cost the same; see the live page price for the current rate. Requests that fail validation are not charged.

Models & Capabilities Krea 3
How many images per request?

One image per request. Submit several tasks, or rerun with different seeds, when you need more.

Models & Capabilities Krea 3
How do I use the negative prompt?

List what you want kept out in negative_prompt, for example people, text, watermark, up to 1024 characters. It applies to berry only: with jam the Playground hides the field and does not send it, and jam ignores it if sent via the API.

Models & Capabilities Krea 3
How do I get reproducible results?

Pass a fixed seed (0–2147483647) and keep the prompt, size, variant and reasoning_level unchanged. For a series, change one detail at a time.

Models & Capabilities Krea 3
How accurate is in-image text?

Short headlines usually come out right. Put the exact words in quotes and describe placement and type style; for text-heavy layouts use berry with reasoning medium.

Models & Capabilities Krea 3
What is the output format?

output_format accepts png (default) or jpeg. Download the image from output[].url once the task succeeds; URLs expire, so copy results to your own storage.

Models & Capabilities Krea 3
Do I need a reference image?

No. It is a text-to-image model and accepts a text description only.

Models & Capabilities Krea 3
Does the same seed give the same image on jam?

No. A seed reproduces results within one variant; berry and jam with the same seed produce different framing and color. To compare them, treat each as a separate option and run both.

Models & Capabilities Krea 3
Which resolutions does LTX 2.5 Fast support?

720p and 1080p, with 720p as the default, for both text-to-video and image-to-video.

Models & Capabilities LTX 2.5 Fast
How long can a clip be?

2 to 20 seconds: 2, 3, 4, 5, 6, 8, 10, 12, 14, 16, 18 or 20, with 6 as the default.

Models & Capabilities LTX 2.5 Fast
Does it support vertical video?

Yes. Set aspect_ratio to 9:16 for vertical output; the default is 16:9 landscape.

Models & Capabilities LTX 2.5 Fast
What is native audio, and does it cost extra?

With generate_audio on (the default) the model also creates ambience, effects or music. Turning it off gives a silent clip at the same price.

Models & Capabilities LTX 2.5 Fast
How is it billed?

Per second of video; the per-second rate depends on resolution, and 1080p costs more than 720p. See the pricing page for current prices.

Models & Capabilities LTX 2.5 Fast
Text-to-video or image-to-video?

Use text-to-video when you only have an idea; use image-to-video when you already have a product shot, poster or storyboard frame and want tighter control of the subject.

Models & Capabilities LTX 2.5 Fast
How does the last frame work?

Pass two images in image_urls; the second is the last frame. The model moves naturally from the first to the last frame — good for day-to-night or open/close transitions.

Models & Capabilities LTX 2.5 Fast
What are the frame image requirements?

JPG, PNG or WebP image URLs. The closer the composition and camera position of the two frames, the smoother the transition.

Models & Capabilities LTX 2.5 Fast
Can I change the frame rate?

It is fixed at 25fps; no parameter is needed.

Models & Capabilities LTX 2.5 Fast
Does image-to-video follow the image's aspect ratio?

The output frame is set by aspect_ratio. Match it to the first frame to avoid cropping or padding.

Models & Capabilities LTX 2.5 Fast
How should I write prompts?

Use subject + action + camera + sound, e.g. "a barista pours latte art, close-up, morning light, soft clinking cups". For image-to-video, describe only the motion to add.

Models & Capabilities LTX 2.5 Fast
How long are results kept?

Result links expire. Download or copy clips to your own storage if you need to keep them.

Models & Capabilities LTX 2.5 Fast
How do I use this entry without an image?

Describe the scene, action, camera and sound. Both frames may be omitted; text-only requests use the default landscape canvas. Existing applications can keep minimax-h3-max. For a new creation with a chosen aspect ratio, use MiniMax H3 Max Text-to-Video.

Models & Capabilities
Can I use an image as a first or last frame?

Supply a first frame, last frame or both. A first frame sets the opening and a last frame guides the ending. The canvas follows the first frame when present, otherwise the last frame. Do not send aspect_ratio with frame images.

Models & Capabilities
Can the legacy ratio field control portrait output?

This entry retains aspect_ratio for compatibility, but the field is not guaranteed to change the generated canvas. Use MiniMax H3 Max Text-to-Video to choose 9:16 or another ratio. With images, their composition determines the canvas.

Models & Capabilities
Do I have to change my existing code?

No. The model name, endpoint, default duration and default resolution are unchanged, and requests that were valid before remain valid. Add the new optional fields only when you want to use them.

Models & Capabilities
How does this entry differ from the new text-to-video and image-to-video entries?

This entry keeps the original behavior: images are optional and text-only requests default to landscape. Text-to-Video accepts text only and lets you choose the frame shape; Image-to-Video needs at least a first or last frame. Duration, resolution, seed, expansion and target audio are written the same on all three.

Models & Capabilities
How long can the prompt be?

Up to 7,000 characters on this entry. The text-to-video and image-to-video entries allow a longer prompt when you need extended shot notes.

Models & Capabilities
Should I switch old requests to 1080P by default?

Leave the default alone. A request that omits resolution still returns 768P, and 1080P is used only when you send it. Try a few tasks first and compare the finished clips before deciding.

Models & Capabilities
Why does the same prompt give a different frame shape with and without an image?

The canvas comes from the image. A first frame sets it when present, a last frame sets it when used alone, and text-only requests use the default landscape. To fix the shape for text-only work, use the text-to-video entry.

Models & Capabilities
Do I need to prepare an image?

No. This entry creates video from text and does not accept first- or last-frame images. Choose MiniMax H3 Max Image-to-Video when an image should set the opening or ending.

Models & Capabilities
How do I choose framing for the destination?

Choose 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16. Landscape suits environments and openers, portrait suits short videos, and square fits feeds. Describe the subject’s position and room for movement so the shot fits the chosen frame.

Models & Capabilities
How do I write a clearer shot description?

Start with the subject and setting, then add one continuous action, camera direction and sound. Avoid crowding a short clip with many events. Use prompt expansion for a fuller description and refine it after reviewing the clip.

Models & Capabilities
Does 9:16 really return a vertical video?

Yes. This entry generates directly in the chosen frame shape, and our 768P test returned 768×1344.

Models & Capabilities
Can I combine target audio with a very short duration?

No. With target audio the video must be at least 2 seconds and the audio itself at least 2 seconds. For clips under a second, let the model generate its own sound.

Models & Capabilities
How do I get several frame shapes of one idea in a similar style?

Keep the prompt and seed fixed and change only aspect_ratio. The same seed does not guarantee identical pictures, but style, color and motion tendencies usually stay closer, which makes selection easier.

Models & Capabilities
When should I turn prompt expansion on?

Use balanced or quality when you only have a short idea and want the model to add detail. Use disabled when the shot, light and action are already written and you want your own wording kept. Expansion does not change the requested length.

Models & Capabilities
Why does the returned length differ slightly from my request?

Video is generated in frames, so the final length lands on the nearest frame boundary. In our tests the difference was under a third of a second. For exact edit points, use the length of the returned video.

Models & Capabilities
What is 1080P good for here?

1080P is an output tier refined from a 768P source and suits final delivery or large-screen playback. For drafting and comparing ideas, 480P or 768P is quicker.

Models & Capabilities
Do I need both a first and last frame?

Supply at least one image. A first frame sets the opening, a last frame alone guides the ending, and both define the opening and ending compositions. These are video endpoint frames, not general style references.

Models & Capabilities
Which image determines the canvas?

The first frame sets the canvas when present; otherwise the last frame does. This entry does not accept aspect_ratio. Prepare a landscape or portrait image for your destination and preferably match both frame shapes. The video determines the final pixel dimensions.

Models & Capabilities
Should the prompt repeat what is in the image?

The image already supplies the subject and composition. Focus the prompt on action, environmental changes and how the camera continues. To guide an ending position, use a last frame with a motion direction and allow enough time for the action.

Models & Capabilities
Is a last frame alone enough?

Yes. A last frame used alone is both the ending picture and the frame shape for the whole clip. Describe the action that leads from the opening to that ending in the prompt.

Models & Capabilities
What happens if the first and last frames have different sizes?

The first frame sets the frame shape and the last frame is handled to match, but a big difference can make the transition feel awkward. Use the same orientation and similar compositions for both.

Models & Capabilities
Can I pass an image to control style rather than start and end?

No. On this entry the images are the video's start and end frames, not general style references. If you want to start from text, use the text-to-video entry.

Models & Capabilities
How do I combine target audio with images?

Send target_audio_url and the finished soundtrack is replaced with your audio. The audio must be at least 2 seconds and no more than 15 MB, and the video at least 2 seconds. It does not guarantee that the picture follows the sound, so describe the action in the prompt.

Models & Capabilities
What does 1080P do for image-to-video?

1080P is an output tier refined from a 768P source. In our test with two portrait frames it returned 1080×1920. Use it for final delivery and use 480P when comparing prompts.

Models & Capabilities
Why try a very short duration first?

Duration accepts fractions from 0.92 to 15 seconds. A 1.5–2 second test confirms that the frame was used correctly and the orientation is what you expect before you scale up to the final length.

Models & Capabilities
How do I switch between generation and editing?

Use a prompt alone to generate an image. Add references and describe the requested changes and details to preserve for editing. Both use the same model ID. Tasks are independent; supply the previous result again to continue editing.

Models & Capabilities
How should I use reference images?

Add up to 10 JPEG, PNG or WebP images. Assign each a subject, setting or style role and list details to preserve. URLs must be publicly accessible; API images must be no larger than 30 MB each.

Models & Capabilities
How does auto choose the aspect ratio?

auto lets the model choose a layout from the prompt and references. Select an explicit ratio for a poster or advertising placement. Resolution tiers do not specify a fixed pixel width and height. Ratios guide composition; actual pixel ratios can vary, especially at 1:8 and 8:1. Check output dimensions before cropping for a fixed placement.

Models & Capabilities
How can I make poster lettering easier to read?

Write each exact phrase, specify the position of the headline, label and subtitle, and exclude extra text. Check the completed image character by character, especially small type and long passages.

Models & Capabilities
Can I request transparency or multiple images?

Each task generates one JPG or PNG image. A transparent-background switch and image-count parameter are not exposed. Choosing PNG does not guarantee transparency.

Models & Capabilities
How do I retrieve and save results?

Save taskId, retrieve GET /v1/tasks/:id until completion, then read output[].url. You can also set callback.url for terminal notifications. Save results promptly if you need to retain them.

Models & Capabilities
What kinds of projects can I create?

Start with fashion portraits and furniture settings, or supply an image for reference editing. This page shows a generated coastal portrait, interior and outfit comparison. Results depend on the input; inspect details before delivery.

Models & Capabilities
How should I choose 1K, 2K or 4K?

Use 1K to explore a direction, 2K to inspect materials, spaces or layout, and consider 4K for larger displays or later cropping. Higher resolution does not automatically fix lettering or structural errors; actual dimensions vary by aspect ratio.

Models & Capabilities
How can I edit while preserving the composition?

Supply the original image and separate changes from details to preserve, such as replacing a jacket while retaining the face, pose and background. Compare garment edges, hands and lighting against the original. Supply the latest result again for another edit.

Models & Capabilities
How can I give images a photographic feel?

Specify the subject, movement, fabric or surface material, camera position and light direction, and limit props. Inspect skin, hands, reflections and shadows. A general request for premium photography or a higher resolution does not guarantee accurate details.

Models & Capabilities
What inputs does OmniHuman 1.5 need?

One portrait image (image_url) and one audio clip (audio_url); both are required. Prompt, masks, resolution, fast mode and seed are optional.

Models & Capabilities
What kind of portrait works best?

A single subject with a clear, unobstructed face gives the steadiest result; photos of people, pets, illustrations and anime characters all work, and half- or full-body images bring out more gestures. The video keeps the image's aspect ratio, so there is no ratio setting.

Models & Capabilities
What are the audio requirements?

Speech or singing in mp3, wav, aac, ogg or m4a, up to 10MB. It must be shorter than 60 seconds, and 15 seconds or less gives the best quality. The video is as long as the audio.

Models & Capabilities
Can I use singing or audio in English and other languages?

Yes. The driving audio can be speech or singing, and lip-sync does not depend on language; give the same portrait voice-overs in different languages to generate a version for each.

Models & Capabilities
Is the prompt required?

No. Without a prompt the model chooses expressions and motion from the audio. Add one of up to 300 characters to steer expression, gestures or camera; Chinese, English, Japanese, Korean, Spanish and Indonesian are supported.

Models & Capabilities
How do I pick the speaker in a group shot?

Upload mask images the same size as the portrait (speaker in white, the rest in black) as mask_urls, up to five; the masked person lip-syncs.

Models & Capabilities
How is it different from HeyGen Avatar IV?

OmniHuman 1.5 is audio-driven only but supports pets and anime characters, produces richer expressions and body motion, and can pick the speaker with masks. HeyGen Avatar IV can work from a script with a preset voice and lets you set the aspect ratio and background padding.

Models & Capabilities
How is OmniHuman 1.5 billed?

By the real seconds of generated video, with 720p and 1080p at the same rate. A hold for the 60-second maximum is taken at creation, so your balance must cover 60 seconds to submit; it is settled to the actual output seconds afterwards with the difference refunded, and failed tasks are refunded in full. See the live price on this page.

Models & Capabilities
What is PixVerse V6 Image to Video?

The image-to-video version of PixVerse V6: upload an image as the first frame, describe the motion, camera and atmosphere to add, and get a 1 to 15 second clip at 360p to 1080p with optional synchronized audio and style presets.

Models & Capabilities PixVerse V6
What determines the output aspect ratio?

The aspect ratio of the image_url first frame, which is why image to video has no aspect_ratio parameter. Upload a vertical image for vertical output and a landscape image for 16:9.

Models & Capabilities PixVerse V6
How should I write the prompt?

Describe only what should change: subject motion, camera movement, environmental change and sound. There is no need to restate what the image already shows. Say what must stay put, such as “camera stays still” or “background unchanged”.

Models & Capabilities PixVerse V6
What are the input image requirements?

image_url must be an image address reachable on the public internet, and the Playground also accepts a direct upload. A clear subject, a complete composition and room for the motion make stable movement more likely.

Models & Capabilities PixVerse V6
Which lengths and resolutions are supported?

duration is any whole number of seconds from 1 to 15, 5 by default; resolution is 360p, 540p, 720p or 1080p, 720p by default.

Models & Capabilities PixVerse V6
Can it generate sound?

Yes. Set generate_audio to true to generate background music, sound effects and dialogue with the video, and describe the sound you want in the prompt. It defaults to false, which returns picture only.

Models & Capabilities PixVerse V6
How do the style presets work?

style accepts anime, 3d_animation, clay, comic or cyberpunk. Leave it out to keep the model's natural look, and pass a value only when you want a consistent stylized result.

Models & Capabilities PixVerse V6
How is the prompt length limit counted?

prompt and negative_prompt are each limited to 2048 bytes in UTF-8. An English letter takes 1 byte and a CJK character takes 3 bytes, so an all-Chinese prompt fits about 680 characters.

Models & Capabilities PixVerse V6
What are negative_prompt and seed for?

negative_prompt lists what to keep out, such as flicker, text watermarks or distorted limbs. A fixed seed lets the same prompt and settings reproduce the same result, which makes it easy to change one variable at a time.

Models & Capabilities PixVerse V6
How is PixVerse V6 billed?

Per second of output, with the rate set by resolution and whether audio is generated, for eight tiers in total: the rate rises from 360p to 1080p, and sound on costs more than sound off. Current rates on the pricing page and your task record show the exact amount.

Models & Capabilities PixVerse V6
How can I keep test runs cheap?

Validate motion and composition with 360p or 540p, a few seconds and audio off, then move to 720p or 1080p with audio for the final version. The examples on this page include both 540p, 4-second test clips and 720p final renders.

Models & Capabilities PixVerse V6
How do I call it through the API?

Submit an asynchronous task to POST /v1/tasks and you get a taskId back immediately. In production pass callback.url to receive the terminal notification; while debugging, poll GET /v1/tasks/:id and download the MP4 from output[].url on success.

Models & Capabilities PixVerse V6
What is PixVerse V6 Text to Video?

The text-to-video version of PixVerse V6: write one prompt and get a 1 to 15 second clip at 360p to 1080p, with optional synchronized audio and five style presets, without uploading any image.

Models & Capabilities PixVerse V6
Which aspect ratios are supported?

aspect_ratio accepts eight values: 16:9, 4:3, 1:1, 3:4, 9:16, 2:3, 3:2 and 21:9, with 16:9 by default. Pick 9:16 for vertical short video or 21:9 for a widescreen look.

Models & Capabilities PixVerse V6
What if I need a fixed opening frame?

Use pixverse-v6/image-to-video instead: upload an image as the first frame and the video starts from it. Text to video suits shots imagined from scratch; image to video suits existing assets or a composition you need to lock.

Models & Capabilities PixVerse V6
What is Qwen Image 2.1 Pro best for?

Balanced text-to-image and image editing with up to 10 reference images and solid Chinese/English text rendering.

Models & Capabilities
How many references are allowed, and does their order matter?

One to ten; omit them for pure text-to-image. References are numbered in the order you send them, so a prompt can address image 1 and image 2 separately — the furniture case on this page gives the subject in image 1 and the scene in image 2, and neither leaks into the other.

Models & Capabilities
How do aspect ratio and pixel size relate?

They are the same setting: each of the five aspect_ratio values maps to one fixed 2K-class pixel size (1:1 to 2048*2048, 16:9 to 2688*1536, 9:16 to 1536*2688, 4:3 to 2368*1728, 3:4 to 1728*2368). Omit it and the model picks a size from your prompt and references. Choose a ratio when layout matters; there is no separate pixel-size control.

Models & Capabilities
Should prompt optimization stay on?

It defaults to on and expands a short description into fuller drawing instructions, which the documentation puts at roughly 3–5 seconds of extra latency. Leave it on when you are starting from one sentence; if your prompt already fixes the image, its text and their positions, the extra detail it adds may not be what you asked for, so turning it off is a reasonable choice.

Models & Capabilities
How is it priced?

Per successfully generated image, at the same rate for text-to-image and reference editing; reference images themselves are not charged. One request returns exactly one image (n is fixed at 1).

Models & Capabilities
How do I choose between this, Qwen Image 2.0 and 3.0?

Pick 2.1 Pro for balanced output quality: generation and editing in one model, five fixed 2K-class ratios, no negative prompt. Choose Qwen Image 2.0 when you need size presets or a negative prompt. For more prompt-driven editing in 1K/2K tiers, try Qwen Image 3.0.

Models & Capabilities
Can it render Chinese text reliably?

The Chinese poster case on this page spells out three lines word by word and comes back with the headline, the date and the venue all correct and nothing extra added. The key is not leaving copywriting to the model: give each line's exact text, its size level and its position, and state that no other text should appear.

Models & Capabilities
Can it produce a transparent background?

The documentation describes RGBA transparent PNG output driven by prompt wording, but in our runs on both tiers the results came back fully opaque (alpha 255 everywhere), so the behaviour is prompt-sensitive and not dependable. Use a dedicated background-removal model when you need a guaranteed transparent cut-out.

Models & Capabilities
Pro or Turbo?

Same inputs and capabilities, different speed, detail and price: Pro for final assets, product shots and precise text layout; Turbo for volume drafts, social assets and direction checks.

Models & Capabilities
What is Qwen Image 2.1 Turbo best for?

High-value, fast generation: concept drafts, social assets, stickers, and volume illustration, with the same capabilities as Pro at a lower price.

Models & Capabilities
Does it accept reference images, and does order matter?

Yes, like Pro it edits with 1–10 references; omit them for text-to-image. References are numbered in the order you send them, so a prompt can address image 1 directly — the photo-to-watercolor case on this page restyles a single reference. When you need both a subject and a scene, assign them as image 1 and image 2 and say which is which.

Models & Capabilities
How do aspect ratio and pixel size relate?

Same as Pro: each of the five aspect_ratio values maps to one fixed 2K-class size (1:1 2048*2048, 16:9 2688*1536, 9:16 1536*2688, 4:3 2368*1728, 3:4 1728*2368), or omit it and let the model decide. For social assets, set 1:1 or 16:9 explicitly so the framing does not follow the reference.

Models & Capabilities
Should prompt optimization stay on?

It defaults to on and expands a single sentence into fuller drawing instructions, which the documentation puts at roughly 3–5 seconds of extra latency. Leave it on for quick drafts; turn it off when your prompt is already specific or you want to isolate one variable in a layout test.

Models & Capabilities
How is it priced?

Per successfully generated image at a flat per-image rate below Pro's; reference images are not charged separately, and one request returns exactly one image (n is fixed at 1).

Models & Capabilities
How do I choose between Turbo, Pro and Qwen Image 2.0?

Turbo and Pro share the same inputs, reference limit, ratios and text rendering: pick Turbo for speed and volume, Pro for finer final images. Choose Qwen Image 2.0 when you need size presets or a negative prompt.

Models & Capabilities
What should I know about restyling?

In our tests Turbo shifted style conservatively: asking only for a watercolor look returned something closer to a weathered photograph. Rewriting the instruction as a redraw — simplified shapes, omitted detail, wet-on-wet edges — produced a clearly painted illustration while keeping the building structure and the furniture layout of the reference. Ask for a redraw and simplification, not just a style name.

Models & Capabilities
How many images per request, and when should I pick Turbo?

One image per request (n is fixed at 1); submit several tasks or rerun with different seeds when you need options. Use Turbo for fast drafts and volume, and Pro for final assets where detail and texture matter most.

Models & Capabilities
What inputs does Ray 3.2 image-to-video need?

Two are required: prompt (1–6000 characters) and the first frame image_url. Optional are end_image_url as the last frame and resolution (540p, 720p or 1080p, default 540p). In the Playground you can upload a local image directly.

Models & Capabilities Ray 3.2
Can I set the video length?

No. Every image-to-video clip is 5 seconds and the request has no duration field. For a 10-second shot, use Ray 3.2 text-to-video, which supports 5 and 10 seconds.

Models & Capabilities Ray 3.2
What decides the output aspect ratio?

The first frame; there is no aspect_ratio field. The output snaps to the closest of 16:9, 9:16, 1:1, 4:3, 3:4 and 21:9, trimming the edges when the ratios differ. The desert example here starts from a roughly 3:2 image (1024×687) and comes back at 4:3, 1104×832; the iced espresso example starts from a 9:16 image and stays 9:16 with virtually nothing trimmed. To control the ratio exactly, crop the first frame to one of those six ratios first.

Models & Capabilities Ray 3.2
How do first and last frames work?

Pass the first frame in image_url and the last in end_image_url; the video moves from one to the other, so the prompt only needs to describe the change in between. The closer the two images are in camera position, light and subject placement, the smoother the transition; when the framing differs the model adds camera movement to bridge it, as in the peony example here.

Models & Capabilities Ray 3.2
Does adding a last frame cost more?

No. Image-to-video is billed per clip by resolution, with or without end_image_url, and because the length is fixed at 5 seconds it is not billed per second either.

Models & Capabilities Ray 3.2
What output sizes do the three resolutions give?

The exact size follows the first frame's ratio. From the three runs here: a 16:9 first frame at 540p came back at 960×540, a roughly 3:2 frame at 720p at 4:3 1104×832, and a 9:16 frame at 1080p at 1080×1920. Check motion and framing at 540p first, then render the final at 720p or 1080p. The iced espresso example here runs at 1080p, and the glass and droplets come back visibly crisper than in the lower-tier examples.

Models & Capabilities Ray 3.2
Does the video have sound?

No. The output is an MP4 without an audio track; add music, narration or sound effects in post.

Models & Capabilities Ray 3.2
How should I prompt image-to-video?

The first frame already sets the subject and scene, so the prompt should focus on what happens next: the subject's action, how the camera moves, and how light or weather changes. Re-describing what is already in the image helps less than saying what must stay the same.

Models & Capabilities Ray 3.2
How is Ray 3.2 image-to-video billed?

Per clip by the selected resolution, with 540p the cheapest and 1080p the highest; failed tasks are refunded. See current rates on the pricing page.

Models & Capabilities Ray 3.2
How do I call it through the API?

Send an asynchronous task to POST /v1/tasks with model set to ray-3.2/image-to-video and prompt and image_url in input, adding end_image_url and resolution as needed. The call returns a taskId right away; receive the result through callback.url or poll GET /v1/tasks/:id.

Models & Capabilities Ray 3.2
Image-to-video or text-to-video?

Use image-to-video when you already have the picture you want (a product shot, a design, a frame from an earlier clip) or need to fix the ending frame. Use text-to-video when you only have an idea in words, need 10 seconds, or want to choose the aspect ratio. The three 5-second tiers cost the same on both routes; the 10-second tiers belong to text-to-video alone.

Models & Capabilities Ray 3.2
What lengths does Ray 3.2 text-to-video support?

5 or 10 seconds, passed as the integer 5 or 10 in duration; the default is 5.

Models & Capabilities Ray 3.2
Which resolutions and aspect ratios are available?

Resolution is 540p, 720p or 1080p (default 540p); aspect ratio is 16:9, 9:16, 1:1, 4:3, 3:4 or 21:9 (default 16:9). The vertical example here used 540p at 9:16 and came back at 540×960; the landscape one used 720p at 16:9 and came back at 1280×720; the 10-second example used 720p at 21:9 and came back at 1680×720.

Models & Capabilities Ray 3.2
How do 10-second and 5-second prices compare?

Billing is per clip, set by both length and resolution. At the same resolution the 10-second tier costs about three times the 5-second tier, so confirm the motion at 5 seconds before rendering a 10-second version. The 10-second example on this page runs at 720p and was billed at about three times the 5-second 720p tier. See current rates on the pricing page.

Models & Capabilities Ray 3.2
Can I upload a reference image or first frame?

Text-to-video takes a text prompt only. To start from an image or also set the last frame, use Ray 3.2 image-to-video (ray-3.2/image-to-video).

Models & Capabilities Ray 3.2
How do I write a prompt that works well?

Prompts can be up to 6000 characters. Write in the order subject and setting, action, camera move, then light and mood, and give the camera a direction and speed such as “low orbit around her” or “slow crane up”. One main action per clip is steadier than packing in several events.

Models & Capabilities Ray 3.2
Is text inside the picture reliable?

No. In the neon dancer example on this page, the lettering on the signs is invented and unreadable. Add brand names, slogans or captions in post.

Models & Capabilities Ray 3.2
Does it work for vertical short video?

Yes. Pick 9:16 to render vertical directly instead of cropping a landscape clip; 1:1 and 3:4 suit feeds, and 21:9 suits widescreen openers.

Models & Capabilities Ray 3.2
How is Ray 3.2 text-to-video billed?

Per clip by the combination of length and resolution, six tiers in all, with 540p at 5 seconds the cheapest; failed tasks are refunded.

Models & Capabilities Ray 3.2
What does Recraft Creative Upscale do?

It enlarges an image to a higher resolution and redraws texture and detail along the way. There is no prompt; you send a single image URL.

Models & Capabilities Recraft Creative Upscale
How is it different from Recraft Crisp Upscale?

Crisp Upscale stays close to the original, which suits text and product shots that must not change. Creative Upscale redraws detail for a richer image, but local structure can differ from the original.

Models & Capabilities Recraft Creative Upscale
Will it change the content?

Overall composition, subject and colors are largely kept, but detail is redrawn. In the example on this page, the cabin's logs, stone wall and railing come back cleaner, the small tree beside the porch is gone, and the stones are arranged differently. Check images with strict detail requirements in the Playground first.

Models & Capabilities Recraft Creative Upscale
What are the input requirements?

One publicly accessible http(s) image URL, up to 20 MB, each side 256–4096 pixels and at most 16 MP in total. The example on this page uses a PNG image, and PNG is the recommended input for now.

Models & Capabilities Recraft Creative Upscale
How large can the output be?

The example on this page goes from 1254×1254 to 4096×4096. Output size for other inputs depends on what the task returns.

Models & Capabilities Recraft Creative Upscale
Do I need a prompt?

No. The model takes a single parameter, image_url, with no prompt, scale or style settings.

Models & Capabilities Recraft Creative Upscale
What format is the output?

Download the image from output[].url in the task result. The example on this page came back as a WebP image.

Models & Capabilities Recraft Creative Upscale
How is it billed?

Per output image, one per request. See the current price above and on the pricing page.

Models & Capabilities Recraft Creative Upscale
Which images is it good for?

Illustrations, landscapes and scenes where redrawn detail is fine, such as refreshing old assets or enlarging smaller generated images for display. Images with exact text, logos or product detail are better suited to Crisp Upscale.

Models & Capabilities Recraft Creative Upscale
How long are results kept?

Result URLs expire, so copy the image to your own storage once the task succeeds.

Models & Capabilities Recraft Creative Upscale
What does Recraft Creative Upscale do?

It enlarges an image to a higher resolution and redraws texture and detail along the way. There is no prompt; you send a single image URL.

Models & Capabilities
How is it different from Recraft Crisp Upscale?

Crisp Upscale stays close to the original, which suits text and product shots that must not change. Creative Upscale redraws detail for a richer image, but local structure can differ from the original.

Models & Capabilities
Will it change the content?

Overall composition, subject and colors are largely kept, but detail is redrawn. In the example on this page, the cabin's logs, stone wall and railing come back cleaner, the small tree beside the porch is gone, and the stones are arranged differently. Check images with strict detail requirements in the Playground first.

Models & Capabilities
What are the input requirements?

One publicly accessible http(s) image URL, up to 20 MB, each side 256–4096 pixels and at most 16 MP in total. The example on this page uses a PNG image, and PNG is the recommended input for now.

Models & Capabilities
How large can the output be?

The example on this page goes from 1254×1254 to 4096×4096. Output size for other inputs depends on what the task returns.

Models & Capabilities
Do I need a prompt?

No. The model takes a single parameter, image_url, with no prompt, scale or style settings.

Models & Capabilities
What format is the output?

Download the image from output[].url in the task result. The example on this page came back as a WebP image.

Models & Capabilities
How is it billed?

Per output image, one per request. See the current price above and on the pricing page.

Models & Capabilities
Which images is it good for?

Illustrations, landscapes and scenes where redrawn detail is fine, such as refreshing old assets or enlarging smaller generated images for display. Images with exact text, logos or product detail are better suited to Crisp Upscale.

Models & Capabilities
How long are results kept?

Result URLs expire, so copy the image to your own storage once the task succeeds.

Models & Capabilities
Does Recraft Crisp Upscale need a prompt?

No. The model takes a single image URL (input.image_url) and no other parameters.

Models & Capabilities
How much does it upscale, and can I choose the factor?

It always doubles width and height; the factor is not configurable. For example 640×640 becomes 1280×1280 and 941×1672 becomes 1882×3344.

Models & Capabilities
What are the input requirements?

JPEG, PNG or WEBP up to 10 MB, as a publicly accessible http(s) URL. You can also upload an image directly in the Playground.

Models & Capabilities
What is the output format?

One JPEG image. PNG or WEBP output cannot be selected at the moment.

Models & Capabilities
Does it change the image content?

No. Crisp Upscale only raises resolution and sharpness; it does not repaint content or invent new text or objects.

Models & Capabilities
Does a transparent PNG stay transparent?

The output is JPEG with no alpha channel. If you need a transparent result, upscale first and then run Recraft Remove Background.

Models & Capabilities
How long does one image take?

Usually between 10 seconds and a minute. Tasks are asynchronous and return a taskId immediately.

Models & Capabilities
Does Recraft Remove Background need a prompt?

No. The model takes a single image URL (input.image_url) with no text parameters, which makes it easy to wire into batch pipelines.

Models & Capabilities
Are hair and fur edges preserved?

Yes. Fine hair, fur, and semi-transparent edges are the model's strength — the example on this page is an API output.

Models & Capabilities
What belongs in the minimum API request?

Submit model plus input.image_url with a publicly accessible https URL.

Models & Capabilities
How is Recraft V4.1 different from V4.1 Flash?

V4.1 focuses on detail and finish; Flash is faster and good for drafts in bulk.

Models & Capabilities Recraft V4.1
Which sizes are supported?

image_size accepts square_hd (default), square, portrait_4_3, portrait_16_9, landscape_4_3 and landscape_16_9.

Models & Capabilities Recraft V4.1
How do I set colors?

Pass colors (an array of RGB objects) and background_color (one RGB object) through the API, each channel 0–255. The online Playground does not show these two fields yet.

Models & Capabilities Recraft V4.1
Are the colors followed strictly?

colors is a preferred palette and the image will lean toward it; if the prompt names clearly conflicting colors, the result may mix both.

Models & Capabilities Recraft V4.1
How many images per request?

One image per request. Submit several tasks for more.

Models & Capabilities Recraft V4.1
What is the output format?

One image, downloaded from output[].url in the task result. Result links expire, so save them promptly.

Models & Capabilities Recraft V4.1
How accurate is in-image text?

Short headlines usually come out right. Put the words in quotes and describe placement and type style for steadier results.

Models & Capabilities Recraft V4.1
Do I need a reference image?

No. It is a text-to-image model and only needs a description.

Models & Capabilities Recraft V4.1
Can I chain Recraft upscale and background removal?

Yes. Pass the output URL to recraft/crisp-upscale for a 2× upscale, or to recraft/remove-background for a transparent background.

Models & Capabilities Recraft V4.1
What changes when switching from Flash to V4.1?

Only the model field: set it to recraft-v4.1/text-to-image and keep everything else.

Models & Capabilities Recraft V4.1
Can I write prompts in Chinese?

Yes, though English prompts are usually steadier for style, material and camera terms; write any in-image text in the language you want it to appear in.

Models & Capabilities Recraft V4.1
How is it billed?

Per output image, one image per request; see the live price for the current rate.

Models & Capabilities Recraft V4.1
What is Recraft V4.1 Flash good for?

Fast photos, flat illustrations, posters and mixed-media images, such as marketing visuals, brand illustration, product shots and concept sketches.

Models & Capabilities
Which sizes are supported?

image_size accepts six presets: square_hd (default), square, portrait_4_3, portrait_16_9, landscape_4_3 and landscape_16_9.

Models & Capabilities
How do I set colors?

Pass colors (an array of RGB objects) and background_color (one RGB object) in the API, each channel 0–255. The online Playground does not show these two fields yet, so use the API for them.

Models & Capabilities
Are the colors applied strictly?

colors strongly guides the palette and the image will mostly stay within it, but it is not a pixel-level guarantee.

Models & Capabilities
How many images per request?

One image per request. Submit several tasks for more.

Models & Capabilities
What is the output format?

One WEBP image, downloaded from output[].url in the task result.

Models & Capabilities
How accurate is in-image text?

Short headlines usually come out right. Put the words in quotes and describe placement and type style for steadier results.

Models & Capabilities
Do I need a reference image?

No. It is a text-to-image model and only needs a description.

Models & Capabilities
Is there a prompt length limit?

prompt accepts 1–10000 characters.

Models & Capabilities
Can I chain Recraft upscale and background removal?

Yes. Pass the output URL to recraft/crisp-upscale for a 2× upscale, or to recraft/remove-background for a transparent PNG.

Models & Capabilities
Is the task synchronous?

No, it is asynchronous. Submitting returns a taskId right away; poll the task or use a callback to get the result.

Models & Capabilities
How is it billed?

Per output image, one per request. See the current price at the top of this page and on the pricing page.

Models & Capabilities
What does Recraft V4 Styles do?

You set a style with 1–10 reference images and describe the content in a prompt; the model draws a new image in that style. The references decide how it looks, the prompt decides what is in it.

Models & Capabilities
Are reference images required, and how many can I send?

Yes. image_urls is required: at least 1 and up to 10 images. The example on this page uses 2 illustrations in the same style.

Models & Capabilities
What formats and sizes can the reference images be?

PNG, JPG or WEBP at publicly accessible http(s) URLs, each under 20 MB and no more than 64 MB in total. SVG is not supported. You can also upload images directly in the Playground.

Models & Capabilities
Will objects from the references show up in the result?

In the example on this page, the teapot, books and bicycle from the references did not appear; the result shows the corner café from the prompt. What carried over was the palette, the flat shading and the hard-edged light. Put the subject you want in the prompt.

Models & Capabilities
How do I choose style_match?

precise keeps the result closer to the reference style, flexible allows more variation; when omitted it behaves as flexible. The example on this page does not set style_match.

Models & Capabilities
Which sizes are supported?

image_size has six presets: square_hd (default), square, portrait_4_3, portrait_16_9, landscape_4_3 and landscape_16_9. The example on this page uses square_hd and came back at 1024×1024.

Models & Capabilities
How is it different from Recraft V4 Styles Pro?

The parameters are the same. Pro returns higher resolution: with square_hd, the Styles example is 1024×1024 and the Pro example is 2048×2048. Their prices differ; see live pricing.

Models & Capabilities
Can I set the palette and background color?

Yes, through the API: colors takes an array of RGB objects and background_color one RGB object, each channel 0–255. The online Playground does not show these two fields yet.

Models & Capabilities
Can I save a style and reuse it?

Saved styles cannot be passed in at the moment. Send the reference images with every request; to keep a set of images consistent, reuse the same references each time.

Models & Capabilities
How many images per request, and in what format?

One image per request, downloaded from output[].url in the task result. The example on this page came back as a WebP image.

Models & Capabilities
How is it billed?

Per output image. The number of references, the size preset and style_match do not change the price. See the current price above and on the pricing page.

Models & Capabilities
How long are results kept?

Result URLs expire, so copy the image to your own storage once the task succeeds.

Models & Capabilities
What does Recraft V4 Styles Pro do?

You set a style with 1–10 reference images and describe the content in a prompt; the model draws a new, higher-resolution image in that style. The references decide how it looks, the prompt decides what is in it.

Models & Capabilities
How high is the output resolution?

The example on this page uses square_hd and came back as a 2048×2048 WebP image. Exact pixel sizes for the other presets depend on what the task returns.

Models & Capabilities
Are reference images required, and how many can I send?

Yes. image_urls is required: at least 1 and up to 10 images. The example on this page uses 2 watercolors in the same style.

Models & Capabilities
What formats and sizes can the reference images be?

PNG, JPG or WEBP at publicly accessible http(s) URLs, each under 20 MB and no more than 64 MB in total. SVG is not supported. You can also upload images directly in the Playground.

Models & Capabilities
Will objects from the references show up in the result?

In the example on this page, the boats, wooden gate and flower beds from the references did not appear; the result shows the cliff lighthouse from the prompt. What carried over was the pale blue sky, the beige stonework and loose watercolor washes with white paper showing. Put the subject you want in the prompt.

Models & Capabilities
How do I choose style_match?

precise keeps the result closer to the reference style, flexible allows more variation; when omitted it behaves as flexible. The example on this page does not set style_match.

Models & Capabilities
Which sizes are supported?

image_size has six presets: square_hd (default), square, portrait_4_3, portrait_16_9, landscape_4_3 and landscape_16_9.

Models & Capabilities
How is it different from Recraft V4 Styles?

The parameters are the same. Pro returns higher resolution: with square_hd, the Styles example is 1024×1024 and the Pro example is 2048×2048. Their prices differ; see live pricing.

Models & Capabilities
Can I set the palette and background color?

Yes, through the API: colors takes an array of RGB objects and background_color one RGB object, each channel 0–255. The online Playground does not show these two fields yet.

Models & Capabilities
Can I save a style and reuse it?

Saved styles cannot be passed in at the moment. Send the reference images with every request; to keep a set of images consistent, reuse the same references each time.

Models & Capabilities
How is it billed?

Per output image, one per request. The number of references, the size preset and style_match do not change the price. See the current price above and on the pricing page.

Models & Capabilities
How long are results kept?

Result URLs expire, so copy the image to your own storage once the task succeeds.

Models & Capabilities
What first-frame image can I use?

One image per request in jpeg, png or webp, up to 10MB; SVG isn't supported. Ideally match the image to the output ratio. When they differ, the first frame is center-cropped, so keep the subject in the middle.

Models & Capabilities Runway Gen 4.5
Can I add a last frame?

No. This model uses a single first frame, and the ending comes from the prompt and the model. To pin down the final picture, Kling 2.5 Turbo Image-to-Video accepts an optional last frame.

Models & Capabilities Runway Gen 4.5
Which ratios, lengths and resolution are supported?

16:9, 9:16, 4:3, 3:4, 1:1, defaulting to 16:9; 5, 8 or 10 seconds in every ratio, defaulting to 5; resolution is fixed at 720p.

Models & Capabilities Runway Gen 4.5
Is there audio in the output?

No. Clips are picture only, so add music and voice-over in your edit.

Models & Capabilities Runway Gen 4.5
What should the prompt cover?

The first frame already sets the scene, so describe motion and camera, for example “clouds drift, slow push-in”, up to 1,000 characters. Give a person one or two actions and the expression stays natural.

Models & Capabilities Runway Gen 4.5
How long are generated videos kept?

Output links last about 7 days by default. Download anything you want to keep, or choose long-term storage when you create the task; past runs stay in your task history.

Models & Capabilities Runway Gen 4.5
Image-to-Video or Runway Gen-4.5 Text-to-Video: which should I use?

Use Image-to-Video when the subject and composition should follow your picture, or when you need 1:1, 4:3 or 3:4. Use Text-to-Video when you start from words; it offers landscape and portrait only, and portrait has no 8-second option.

Models & Capabilities Runway Gen 4.5
Which aspect ratios, lengths and resolution are supported?

16:9 landscape and 9:16 portrait. Landscape runs 5, 8 or 10 seconds and portrait 5 or 10, defaulting to 5. Resolution is fixed at 720p.

Models & Capabilities Runway Gen 4.5
Why can't I pick 8 seconds in portrait?

That limit comes from the model: 9:16 supports only 5 and 10 seconds. If you need something close to eight seconds vertically, generate ten and trim it in the edit.

Models & Capabilities Runway Gen 4.5
Do the clips have sound?

No. Output is picture only, so add voice, music and effects when you edit. For clips with sound built in, see Kling 2.5 Turbo Text-to-Video.

Models & Capabilities Runway Gen 4.5
How should I write the prompt?

One or two sentences covering subject, setting, action and camera, then time of day, weather and light, up to 1,000 characters. Give the subject one main action; asking it to turn, run and wave at once tends to produce movements that fight each other.

Models & Capabilities Runway Gen 4.5
What if I need a square or 4:3 frame?

Text-to-video offers only 16:9 and 9:16. For 1:1, 4:3 or 3:4, use Runway Gen-4.5 Image-to-Video, which opens on an image and supports all five ratios.

Models & Capabilities Runway Gen 4.5
Can I try it before using the API?

Yes. Sign in, write a description in the Playground at the top of this page and generate five seconds to check the direction. When it works, send the same description through the API.

Models & Capabilities Runway Gen 4.5
Does Seedance 2.5 image-to-video require both a first and last frame?

No. In endpoint-frame mode, first_frame_url alone extends the source into a complete shot; add last_frame_url when the ending state must be controlled. A last frame cannot be used by itself. You can instead use reference_image_urls or reference_audio_urls as a separate reference mode without endpoint frames.

Models & Capabilities Seedance 2.5
Can first/last frames be mixed with multimodal references in one request?

The current HiAPI Playground treats them as separate input modes: use first/last frames, or use reference_image_urls and reference_audio_urls. Keep one mode to avoid conflicting guidance. Use seedance-2.5/reference-to-video when a reference video must guide motion or camera movement.

Models & Capabilities Seedance 2.5
Which durations and resolutions does Seedance 2.5 image-to-video support?

The current request contract accepts integer durations from 4 to 30 seconds at 720p or 1080p. aspect_ratio is adaptive and follows the first frame. The four cases on this page cover 720p at four and five seconds; 30 seconds is a supported contract value with no page example yet.

Models & Capabilities Seedance 2.5
How is Seedance 2.5 image-to-video priced?

Billing scales with output duration and resolution, with separate per-second tiers for 720p and 1080p. Check the live model price and task record for the final charge.

Models & Capabilities Seedance 2.5
How should I prompt a first/last-frame transition?

Describe the continuous change between the endpoints, then lock camera behavior, lighting, and details that must remain stable. A day-to-night shot can specify time passing and lights switching on; a growth shot can define the growth order, locked camera, and no cuts. The images define the endpoints while the prompt directs the path between them.

Models & Capabilities Seedance 2.5
Can Seedance 2.5 generate synchronized audio?

Yes. generate_audio defaults to true and can create synchronized dialogue, ambience, sound effects, and background audio. Set it to false for a silent result. Review volume, dialogue accuracy, and music rights before publishing commercial work.

Models & Capabilities Seedance 2.5
How do I call the Seedance 2.5 image-to-video API?

Send an asynchronous task to POST /v1/tasks, set model to seedance-2.5/image-to-video, and pass prompt, first_frame_url, an optional last_frame_url, resolution, duration, and generate_audio in input. The create request returns a taskId. See the API docs for callbacks, retrieval, and complete media constraints.

Models & Capabilities Seedance 2.5
Why does Seedance 2.5 reference-to-video require a reference video?

The model is designed to extract motion, camera language, timing, or performance from video, so reference_video_urls must contain at least one clip. Images and audio can add identity, style, or sound guidance, but cannot replace the required video.

Models & Capabilities Seedance 2.5
How is reference-to-video different from image-to-video?

Reference-to-video transfers motion, camera language, and rhythm from a clip without promising an exact first or last frame. Use seedance-2.5/image-to-video to extend one image or lock endpoint frames. Reference-to-video is not video editing or extension.

Models & Capabilities Seedance 2.5
How do @video1, @image1, and @audio1 map to uploaded media?

Each media array is numbered independently in upload order. The first reference video is @video1, the first image is @image1, and the first audio clip is @audio1. State whether each item contributes motion, identity, style, or sound.

Models & Capabilities Seedance 2.5
How many references does Seedance 2.5 reference-to-video accept?

It accepts one to ten reference videos, up to thirty images, and up to ten audio clips. Reference videos can total thirty seconds and audio references can separately total thirty seconds. Each video must be 2–30 seconds, 24–60 fps, and at least 409600 pixels.

Models & Capabilities Seedance 2.5
How is Seedance 2.5 reference-to-video priced?

Both reference and generated videos are billed by duration. Cost = price per second × (reference-video duration + output-video duration)

Models & Capabilities Seedance 2.5
Can Seedance 2.5 reference-to-video generate synchronized audio?

Yes. generate_audio defaults to true and can produce synchronized dialogue, effects, and background audio. Set it to false for video-only output. An audio reference can guide sound or rhythm but still requires at least one video reference.

Models & Capabilities Seedance 2.5
How do I call the Seedance 2.5 reference-to-video API?

POST an asynchronous task to /v1/tasks with model at the request root set to seedance-2.5/reference-to-video. Put prompt and reference_video_urls in input, then add optional image, audio, resolution, duration, and output settings. The create request returns a taskId.

Models & Capabilities Seedance 2.5
What is the difference between Seedream 4.5 image-to-image and text-to-image?

The image-to-image endpoint uses uploaded images as editing references for recoloring, scene changes, relighting, style transfer, and multi-image composition. Text-to-image starts from a written prompt without a source image. Use this image-to-image model when subjects, composition, or in-image text should remain recognizable.

Models & Capabilities Seedream 4.5
How many reference images does Seedream 4.5 support?

Each request accepts 1–14 JPEG, PNG, or WebP images up to 10 MB each; SVG is not supported. For multi-image jobs, refer to inputs as “image 1” and “image 2” in the prompt, matching their order in the image_urls array.

Models & Capabilities Seedream 4.5
How is Seedream 4.5 image editing priced, and do more references cost extra?

Billing is per successful output image, not per reference image. The current public rate is the same for 2K and 4K output. See the live model price and task record for the final charge.

Models & Capabilities Seedream 4.5
How do I preserve a subject, composition, or text during an edit?

State both what should change and what must remain unchanged, such as the subject silhouette, camera angle, contact shadow, storefront layout, or calligraphy. Review the result in the Playground first. Keep human review for brand text or layouts that must be exact.

Models & Capabilities Seedream 4.5
When should I choose Seedream 4.5 instead of FLUX.2 [pro] Image to Image?

Seedream 4.5 accepts up to 14 references and is a strong fit for multi-image composition, Chinese scenes, and edits that preserve in-image text. FLUX.2 [pro] Image to Image accepts up to 8 references and provides an alternative for photoreal consistency and editing style. Test both with the same assets when the choice matters.

Models & Capabilities Seedream 4.5
How do I call the Seedream 4.5 image-to-image API?

Send a POST request to /v1/tasks, set model to seedream-4.5/image-to-image, and pass reference images in input.image_urls. The create request returns a taskId while generation continues asynchronously. See the API docs for task retrieval, callbacks, errors, and the full parameter contract.

Models & Capabilities Seedream 4.5
Can I generate and edit with the same model?

Omit references for text-to-image generation. Add one image for editing or several to combine subjects, scenes or styles. Both use the same model ID, with up to 10 references.

Models & Capabilities
How do I control resolution and aspect ratio?

Set size to 1K, 1.5K, 2K or WIDTHxHEIGHT. Use pixel dimensions for an exact ratio: 921600–4624220 total pixels and a ratio between 1:16 and 16:1. A 4K preset is not available.

Models & Capabilities
What does transparent background require?

Use transparent only when editing one reference that already has an alpha channel. Choose PNG or leave the format automatic. It is not a general background-removal switch.

Models & Capabilities
Can I request batches or streamed images?

Each task returns one completed image through task retrieval. Batch generation, streamed images and layer decomposition are not exposed.

Models & Capabilities
How do I retrieve the image?

Save taskId after submission, retrieve the task until success, then read the image URL from output. Follow expireAt for retention and save assets you need to keep.

Models & Capabilities
Can I use these images as accurate product photography?

Text generation is useful for exploring staging, backgrounds and light, or making visual proposals. If an image must match a real product, supply a reference and check structure, color, markings and lettering. Generated concept details are not product facts.

Models & Capabilities
How should I combine several references?

Assign a role to each image: for example, the first supplies the product, the second the setting, and the third only the color palette. Describe the intended result and details to preserve, avoiding conflicting requirements across references.

Models & Capabilities
Can I continue editing a previous result?

Supply the previous result as a reference in a new task and describe the next change. Tasks are independent and do not automatically remember earlier conversations, images or edits.

Models & Capabilities
Where can I check the generation cost?

Check the Playground estimate before running and the task record afterward. Each task produces one image; the page’s pricing section shows the applicable charge.

Models & Capabilities
How many references does Seedream 5.0 Lite Image to Image accept?

It accepts 1 to 14 JPEG, PNG, or WebP images under 10MB each. Use one for focused edits or multiple references for subject and scene composites.

Models & Capabilities
Do more references cost more?

No. Pricing is fixed per output image, independent of reference count, 2K or 4K tier, and aspect ratio. See the live page price for the current amount.

Models & Capabilities
How do I limit the edit to one area?

Describe both the target change and the composition, text, material, lighting, and positions that must remain unchanged.

Models & Capabilities
Can it rewrite text inside an image?

Yes. State the old and new copy and ask it to preserve the original type style, size, position, and glow or print treatment.

Models & Capabilities
How do I submit references through the API?

Pass 1 to 14 public image URLs in input.image_urls when posting to /v1/tasks, together with the prompt, ratio, and 2K or 4K resolution.

Models & Capabilities
What is Seedream 5.0 Lite Text to Image best for?

It is a good fit for text-heavy posters, menus, infographics, marketing banners, and fast multi-ratio delivery. This route accepts text prompts only.

Models & Capabilities
Do 2K and 4K cost the same?

Yes. Both resolutions use the same per-image tier. Aspect ratio and prompt length do not change the price; see the live page price for the current amount.

Models & Capabilities
Which aspect ratios are supported?

It supports 1:1, 4:3, 3:4, 16:9, 9:16, 2:3, 3:2, and 21:9.

Models & Capabilities
Can I upload reference images?

No. Use the separate seedream-5.0-lite/image-to-image model ID for instruction edits or multi-reference composites.

Models & Capabilities
Which fields are required in the minimal API request?

POST /v1/tasks with model plus input.prompt, input.aspect_ratio, and input.resolution. Resolution must be 2K or 4K.

Models & Capabilities
How many references does Seedream 5.0 Pro Image to Image accept?

It accepts 1 to 10 JPEG, PNG, or WebP images under 10MB each. SVG is not supported. Use one image for focused edits or multiple images for subject, style, and scene blending.

Models & Capabilities
How are 1K and 2K priced?

1K and 2K use separate per-output tiers regardless of the number of references. See the live page price for current amounts.

Models & Capabilities
How do I preserve everything outside the edit?

State both the requested change and the exact structure, text, camera, lighting, and object positions that must remain unchanged.

Models & Capabilities
Do generation and editing use the same model ID?

No. This page uses seedream-5.0-pro/image-to-image. Use the separate seedream-5.0-pro/text-to-image ID to generate from text only.

Models & Capabilities
How do I submit reference images through the API?

Pass 1 to 10 public image URLs in input.image_urls together with the prompt, aspect ratio, resolution, and output format to POST /v1/tasks.

Models & Capabilities
What is Seedream 5.0 Pro Text to Image best for?

It is suited to product campaigns, portraits, posters, in-image text, and detailed environments. This route does not accept reference images; use the separate Image to Image model ID for edits.

Models & Capabilities
Which aspect ratios and formats are supported?

It supports 1:1, 4:3, 3:4, 16:9, 9:16, 2:3, 3:2, and 21:9, with PNG or JPEG output.

Models & Capabilities
How should I prompt for text inside an image?

Put the exact copy in quotation marks and specify its position and type style. Validate the layout at 1K, then render the final version at 2K.

Models & Capabilities
How do I call the API?

Submit the model ID, prompt, aspect ratio, resolution, and output format to POST /v1/tasks, then retrieve the asynchronous result with the returned taskId.

Models & Capabilities
Does Kling 3.0 Turbo Image to Video require an input image?

Yes. kling-3.0-turbo/image-to-video is an image-to-video route. The image establishes the subject and opening composition; the prompt directs motion, camera, and environmental change.

Models & Capabilities Kling 3.0 Turbo
Which durations and resolutions does Kling 3.0 Turbo Image to Video support?

The current HiAPI contract supports 3–15s at 720p or 1080p. Check the page controls and API docs for the current options.

Models & Capabilities Kling 3.0 Turbo
How is Kling 3.0 Turbo Image to Video billed?

Billing follows output duration and the selected tier. The page follows the global display unit, while the successful task record contains the final charge.

Models & Capabilities Kling 3.0 Turbo
How can I improve subject consistency?

Use a clear, unobstructed source with a complete composition. Describe the main action and camera first, then name the appearance, material, and background relationships that must remain stable. Validate a short clip first.

Models & Capabilities Kling 3.0 Turbo
Does Hailuo 2.3 Fast Image to Video require an input image?

Yes. hailuo-2.3-fast/image-to-video is an image-to-video route. The image establishes the subject and opening composition; the prompt directs motion, camera, and environmental change.

Models & Capabilities
Which durations and resolutions does Hailuo 2.3 Fast Image to Video support?

The current HiAPI contract supports 6s / 10s. Check the page controls and API docs for the current options.

Models & Capabilities
How is Hailuo 2.3 Fast Image to Video billed?

Each generated video is billed at the selected duration tier. The page follows the global display unit, while the successful task record contains the final charge.

Models & Capabilities
How can I improve subject consistency?

Use a clear, unobstructed source with a complete composition. Describe the main action and camera first, then name the appearance, material, and background relationships that must remain stable. Validate a short clip first.

Models & Capabilities
Does Kling 3.0 Image-to-Video require an input image?

Yes. kling-3.0/image-to-video is an image-to-video route. The image establishes the subject and opening composition; the prompt directs motion, camera, and environmental change.

Models & Capabilities Kling 3.0
Which durations and resolutions does Kling 3.0 Image-to-Video support?

The current HiAPI contract supports 3–15s at 720p or 1080p or 4K. Check the page controls and API docs for the current options.

Models & Capabilities Kling 3.0
How is Kling 3.0 Image-to-Video billed?

Billing follows output duration and the selected tier. The page follows the global display unit, while the successful task record contains the final charge.

Models & Capabilities Kling 3.0
How can I improve subject consistency?

Use a clear, unobstructed source with a complete composition. Describe the main action and camera first, then name the appearance, material, and background relationships that must remain stable. Validate a short clip first.

Models & Capabilities Kling 3.0
Does Grok Imagine Image to Video require an input image?

Yes. grok-imagine/image-to-video is an image-to-video route. The image establishes the subject and opening composition; the prompt directs motion, camera, and environmental change.

Models & Capabilities Grok Imagine
Which durations and resolutions does Grok Imagine Image to Video support?

The current HiAPI contract supports 6–30s at 480p or 720p. Check the page controls and API docs for the current options.

Models & Capabilities Grok Imagine
How is Grok Imagine Image to Video billed?

Billing follows output duration and the selected tier. The page follows the global display unit, while the successful task record contains the final charge.

Models & Capabilities Grok Imagine
How can I improve subject consistency?

Use a clear, unobstructed source with a complete composition. Describe the main action and camera first, then name the appearance, material, and background relationships that must remain stable. Validate a short clip first.

Models & Capabilities Grok Imagine
Does Wan 2.7 Image to Video require an input image?

Yes. wan2.7-video/image-to-video is an image-to-video route. The image establishes the subject and opening composition; the prompt directs motion, camera, and environmental change.

Models & Capabilities Wan 2.7 Video
Which durations and resolutions does Wan 2.7 Image to Video support?

The current HiAPI contract supports 2–15s at 720P or 1080P. Check the page controls and API docs for the current options.

Models & Capabilities Wan 2.7 Video
How is Wan 2.7 Image to Video billed?

Billing follows output duration and the selected tier. The page follows the global display unit, while the successful task record contains the final charge.

Models & Capabilities Wan 2.7 Video
How can I improve subject consistency?

Use a clear, unobstructed source with a complete composition. Describe the main action and camera first, then name the appearance, material, and background relationships that must remain stable. Validate a short clip first.

Models & Capabilities Wan 2.7 Video
Does Grok Imagine 1.5 Image to Video require an input image?

Yes. grok-imagine-1.5/image-to-video is an image-to-video route. The image establishes the subject and opening composition; the prompt directs motion, camera, and environmental change.

Models & Capabilities Grok Imagine 1.5
Which durations and resolutions does Grok Imagine 1.5 Image to Video support?

The current HiAPI contract supports 1–15s at 480p or 720p. Check the page controls and API docs for the current options.

Models & Capabilities Grok Imagine 1.5
How is Grok Imagine 1.5 Image to Video billed?

Billing follows output duration and the selected tier. The page follows the global display unit, while the successful task record contains the final charge.

Models & Capabilities Grok Imagine 1.5
How can I improve subject consistency?

Use a clear, unobstructed source with a complete composition. Describe the main action and camera first, then name the appearance, material, and background relationships that must remain stable. Validate a short clip first.

Models & Capabilities Grok Imagine 1.5
Does HappyHorse 1.1 Reference to Video require an input image?

Yes. happyhorse-1.1/reference-to-video is a reference-to-video route. Reference images constrain subjects, style, and scene details while the prompt directs motion, camera, and environmental change.

Models & Capabilities HappyHorse 1.1
Which durations and resolutions does HappyHorse 1.1 Reference to Video support?

The current HiAPI contract supports 3–15s at 720p or 1080p. Check the page controls and API docs for the current options.

Models & Capabilities HappyHorse 1.1
How is HappyHorse 1.1 Reference to Video billed?

Billing follows output duration and the selected tier. The page follows the global display unit, while the successful task record contains the final charge.

Models & Capabilities HappyHorse 1.1
How can I improve subject consistency?

Use a clear, unobstructed source with a complete composition. Describe the main action and camera first, then name the appearance, material, and background relationships that must remain stable. Validate a short clip first.

Models & Capabilities HappyHorse 1.1
Does Hailuo 2.3 Image to Video require an input image?

Yes. hailuo-2.3/image-to-video is an image-to-video route. The image establishes the subject and opening composition; the prompt directs motion, camera, and environmental change.

Models & Capabilities Hailuo 2.3
Which durations and resolutions does Hailuo 2.3 Image to Video support?

The current HiAPI contract supports 6s / 10s. Check the page controls and API docs for the current options.

Models & Capabilities Hailuo 2.3
How is Hailuo 2.3 Image to Video billed?

Each generated video is billed at the selected duration tier. The page follows the global display unit, while the successful task record contains the final charge.

Models & Capabilities Hailuo 2.3
How can I improve subject consistency?

Use a clear, unobstructed source with a complete composition. Describe the main action and camera first, then name the appearance, material, and background relationships that must remain stable. Validate a short clip first.

Models & Capabilities Hailuo 2.3
Does Veo 3.1 Fast Image to Video require an input image?

Yes. veo-3.1-fast/image-to-video is an image-to-video route. The image establishes the subject and opening composition; the prompt directs motion, camera, and environmental change.

Models & Capabilities Veo 3.1 Fast
Which durations and resolutions does Veo 3.1 Fast Image to Video support?

The current HiAPI contract supports 4s / 6s / 8s at 720p or 1080p or 4k. Check the page controls and API docs for the current options.

Models & Capabilities Veo 3.1 Fast
How is Veo 3.1 Fast Image to Video billed?

Billing follows output duration and the selected tier. The page follows the global display unit, while the successful task record contains the final charge.

Models & Capabilities Veo 3.1 Fast
How can I improve subject consistency?

Use a clear, unobstructed source with a complete composition. Describe the main action and camera first, then name the appearance, material, and background relationships that must remain stable. Validate a short clip first.

Models & Capabilities Veo 3.1 Fast
Does HappyHorse 1.1 Image to Video require an input image?

Yes. happyhorse-1.1/image-to-video is an image-to-video route. The image establishes the subject and opening composition; the prompt directs motion, camera, and environmental change.

Models & Capabilities HappyHorse 1.1
Which durations and resolutions does HappyHorse 1.1 Image to Video support?

The current HiAPI contract supports 3–15s at 720p or 1080p. Check the page controls and API docs for the current options.

Models & Capabilities HappyHorse 1.1
How is HappyHorse 1.1 Image to Video billed?

Billing follows output duration and the selected tier. The page follows the global display unit, while the successful task record contains the final charge.

Models & Capabilities HappyHorse 1.1
How many references does FLUX.2 [pro] Image to Image accept?

It accepts 1 to 8 JPEG, PNG, or WebP references under 10MB each. Assign each image a clear role in multi-image prompts.

Models & Capabilities FLUX.2
What does the auto ratio do?

auto follows the first reference image. Choose an explicit ratio for a fixed landscape, portrait, or square output.

Models & Capabilities FLUX.2
How are 1K and 2K billed?

1K and 2K use separate per-image tiers, while reference count adds no separate fee. See the live page price for current amounts.

Models & Capabilities FLUX.2
How do I reduce subject or logo drift?

Assign each reference a role and state the subject, text, material, position, and lighting that must remain unchanged.

Models & Capabilities FLUX.2
Which API fields are required?

Submit model plus input.prompt, input.image_urls, input.aspect_ratio, and input.resolution.

Models & Capabilities FLUX.2
How many references does Grok Imagine Image to Image accept?

The current public schema accepts 1 to 3 references. The page and API follow that contract.

Models & Capabilities Grok Imagine
How does standard editing differ from Quality?

Standard emphasizes a low flat price and rapid iteration. Quality prioritizes subject consistency, material, and final detail.

Models & Capabilities Grok Imagine
How does auto ratio work?

auto follows the first reference, or you can select one of fourteen landscape, square, and portrait ratios.

Models & Capabilities Grok Imagine
Do more references add a fee?

Yes. Each reference adds a fee on top of the standard output tier; see the live page price for current amounts.

Models & Capabilities Grok Imagine
What belongs in the minimum API request?

Submit model, input.prompt, and input.image_urls. Ratio, resolution, and output format may use defaults.

Models & Capabilities Grok Imagine
How many references does Grok Imagine Quality Image to Image accept?

The current public schema accepts 1 to 3 references. The page and API follow that contract.

Models & Capabilities Grok Imagine Quality
How does Quality differ from standard editing?

Quality prioritizes subject consistency, material, and detail with tiered 1k/2k pricing. Standard is better for low-cost iteration.

Models & Capabilities Grok Imagine Quality
How does auto ratio work?

auto follows the first reference, or you can select one of fourteen landscape, square, and portrait ratios.

Models & Capabilities Grok Imagine Quality
Do more references add a fee?

Yes. Each reference adds a fee on top of the 1k or 2k output tier; see the live page price for current amounts.

Models & Capabilities Grok Imagine Quality
What belongs in the minimum API request?

Submit model, input.prompt, and input.image_urls. Ratio, resolution, and output format may use defaults.

Models & Capabilities Grok Imagine Quality
How many references does Qwen Image 3.0 Image to Image accept?

It accepts 1 to 3 reference images. At least one is required in input.image_urls.

Models & Capabilities Qwen Image 3.0
How does standard editing differ from Pro?

Standard prioritizes lower cost and rapid iteration. Pro is better for exact text replacement, complex materials, and final layouts.

Models & Capabilities Qwen Image 3.0
How are 1K, 2K, and reference images billed?

1K and 2K use the same output tier, and each reference adds a fee. See the live page price for current amounts.

Models & Capabilities Qwen Image 3.0
How do I reduce subject, layout, or text drift?

Separate what should change from the subject, text, position, material, camera, and lighting that must remain fixed. Put exact replacement text in quotation marks.

Models & Capabilities Qwen Image 3.0
What belongs in the minimum API request?

Submit model, input.prompt, and input.image_urls. Ratio, resolution, output format, and prompt extension may use defaults.

Models & Capabilities Qwen Image 3.0
How many references does Qwen Image 3.0 Pro Image to Image accept?

It accepts 1 to 3 reference images. At least one is required in input.image_urls.

Models & Capabilities
How does Pro differ from standard editing?

Pro is better for exact text replacement, complex materials, and final layouts. Standard is better for lower-cost product recoloring, cleanup, and rapid iterations.

Models & Capabilities
How are 1K, 2K, and reference images billed?

1K and 2K use separate output tiers, and each reference adds a fee. See the live page price for current amounts.

Models & Capabilities
How do I reduce subject, layout, or text drift?

Separate what should change from the subject, text, position, material, camera, and lighting that must remain fixed. Put exact replacement text in quotation marks.

Models & Capabilities
What belongs in the minimum API request?

Submit model, input.prompt, and input.image_urls. Ratio, resolution, output format, and prompt extension may use defaults.

Models & Capabilities
What is FLUX1.1 [pro] best for?

Generate polished photography, architecture, food, nature, and commercial visuals with custom sizing and prompt upsampling controls.

Models & Capabilities FLUX 1.1 Pro
Does FLUX1.1 [pro] accept reference images?

No. This route is text-to-image only.

Models & Capabilities FLUX 1.1 Pro
Which output settings are supported?

The page exposes Custom width / height and Custom, 1:1, 16:9, 9:16; use the Playground contract for the final available set.

Models & Capabilities FLUX 1.1 Pro
How is FLUX1.1 [pro] billed?

Billing is per successful generated image. Resolution or quality differences, when present, are reflected by the live page price.

Models & Capabilities FLUX 1.1 Pro
How do I call the FLUX1.1 [pro] API?

POST an asynchronous task to /v1/tasks with model set to flux-1.1-pro and put generation fields inside input.

Models & Capabilities FLUX 1.1 Pro
What is Qwen Image 3.0 best for?

Create typography-led posters, editorial scenes, and product visuals with Chinese and English text, prompt extension, and 1K/2K output.

Models & Capabilities Qwen Image 3.0
Does Qwen Image 3.0 accept reference images?

No. This route is text-to-image only.

Models & Capabilities Qwen Image 3.0
Which output settings are supported?

The page exposes 1K, 2K and 1:1, 16:9, 9:16, 4:3, 3:4; use the Playground contract for the final available set.

Models & Capabilities Qwen Image 3.0
How is Qwen Image 3.0 billed?

Billing is per successful generated image. Resolution or quality differences, when present, are reflected by the live page price.

Models & Capabilities Qwen Image 3.0
How do I call the Qwen Image 3.0 API?

POST an asynchronous task to /v1/tasks with model set to qwen-image-3.0/text-to-image and put generation fields inside input.

Models & Capabilities Qwen Image 3.0
What is Qwen Image 3.0 Pro best for?

Use the Pro route for denser typography, architecture, science editorials, luxury products, and final campaign layouts.

Models & Capabilities
Does Qwen Image 3.0 Pro accept reference images?

No. This route is text-to-image only.

Models & Capabilities
Which output settings are supported?

The page exposes 1K, 2K and 1:1, 16:9, 9:16, 4:3, 3:4; use the Playground contract for the final available set.

Models & Capabilities
How is Qwen Image 3.0 Pro billed?

Billing is per successful generated image. Resolution or quality differences, when present, are reflected by the live page price.

Models & Capabilities
How do I call the Qwen Image 3.0 Pro API?

POST an asynchronous task to /v1/tasks with model set to qwen-image-3.0-pro/text-to-image and put generation fields inside input.

Models & Capabilities
What is GPT Image 2 Image to Image best for?

Edit one or more references with prompt-driven background, lighting, composition, and style changes while preserving source structure.

Models & Capabilities GPT Image 2
Does GPT Image 2 Image to Image accept reference images?

Yes. This route requires reference input and accepts up to 6 images.

Models & Capabilities GPT Image 2
Which output settings are supported?

The page exposes 1K, 2K, 4K and auto, 1:1, 9:16, 16:9, 4:3, 3:4; use the Playground contract for the final available set.

Models & Capabilities GPT Image 2
How is GPT Image 2 Image to Image billed?

Billing is per successful generated image. Resolution or quality differences, when present, are reflected by the live page price.

Models & Capabilities GPT Image 2
How do I call the GPT Image 2 Image to Image API?

POST an asynchronous task to /v1/tasks with model set to gpt-image-2/image-to-image and put generation fields inside input.

Models & Capabilities GPT Image 2
What is FLUX.2 [klein] 4B Image to Image best for?

A compact low-cost editing route for commercial refinements, background changes, recoloring, and rapid reference-based iteration.

Models & Capabilities FLUX.2 Klein 4B
Does FLUX.2 [klein] 4B Image to Image accept reference images?

Yes. This route requires reference input and accepts up to 1 image.

Models & Capabilities FLUX.2 Klein 4B
Which output settings are supported?

The page exposes 0.25 MP, 0.5 MP, 1 MP, 2 MP, 4 MP and 1:1, 16:9, 9:16, 4:3, 3:4; use the Playground contract for the final available set.

Models & Capabilities FLUX.2 Klein 4B
How is FLUX.2 [klein] 4B Image to Image billed?

Billing is per successful generated image. Resolution or quality differences, when present, are reflected by the live page price.

Models & Capabilities FLUX.2 Klein 4B
How do I call the FLUX.2 [klein] 4B Image to Image API?

POST an asynchronous task to /v1/tasks with model set to flux-2-klein-4b/image-to-image and put generation fields inside input.

Models & Capabilities FLUX.2 Klein 4B
What is Nano Banana Pro best for?

Generate high-detail commercial images or remix optional references with broad ratios and up to 4K output.

Models & Capabilities Nano Banana Pro
Does Nano Banana Pro accept reference images?

Yes, up to 8 optional references. Omit them for text-only generation.

Models & Capabilities Nano Banana Pro
Which output settings are supported?

The page exposes 1K, 2K, 4K and auto, 1:1, 16:9, 9:16, 3:4, 4:3, 3:2, 2:3, 5:4, 4:5, 21:9; use the Playground contract for the final available set.

Models & Capabilities Nano Banana Pro
How is Nano Banana Pro billed?

Billing is per successful generated image. Resolution or quality differences, when present, are reflected by the live page price.

Models & Capabilities Nano Banana Pro
How do I call the Nano Banana Pro API?

POST an asynchronous task to /v1/tasks with model set to nano-banana-pro and put generation fields inside input.

Models & Capabilities Nano Banana Pro
What is FLUX.1 [schnell] best for?

A fast, low-cost 1-4 step model for drafts, batch generation, and rapid prompt exploration.

Models & Capabilities FLUX.1 Schnell
Does FLUX.1 [schnell] accept reference images?

No. This route is text-to-image only.

Models & Capabilities FLUX.1 Schnell
Which output settings are supported?

The page exposes 1-12 inference steps and 1:1, 16:9, 9:16, 4:3, 3:4; use the Playground contract for the final available set.

Models & Capabilities FLUX.1 Schnell
How is FLUX.1 [schnell] billed?

Billing is per successful generated image. Resolution or quality differences, when present, are reflected by the live page price.

Models & Capabilities FLUX.1 Schnell
How do I call the FLUX.1 [schnell] API?

POST an asynchronous task to /v1/tasks with model set to flux-schnell/text-to-image and put generation fields inside input.

Models & Capabilities FLUX.1 Schnell
What is Nano Banana 2 Lite best for?

A fixed-1K entry route for high-volume drafts, fast iteration, and optional multi-reference remixing.

Models & Capabilities Nano Banana 2 Lite
Does Nano Banana 2 Lite accept reference images?

Yes, up to 10 optional references. Omit them for text-only generation.

Models & Capabilities Nano Banana 2 Lite
Which output settings are supported?

The page exposes 1K and auto, 1:1, 16:9, 9:16, 4:3, 3:4, 8:1, 1:8; use the Playground contract for the final available set.

Models & Capabilities Nano Banana 2 Lite
How is Nano Banana 2 Lite billed?

Billing is per successful generated image. Resolution or quality differences, when present, are reflected by the live page price.

Models & Capabilities Nano Banana 2 Lite
How do I call the Nano Banana 2 Lite API?

POST an asynchronous task to /v1/tasks with model set to Nano-Banana-2-Lite and put generation fields inside input.

Models & Capabilities Nano Banana 2 Lite
What is Grok Image 2.0 Text to Image best for?

Generate product, editorial, landscape, and character visuals in low or medium quality at 1K/2K across square to ultrawide formats.

Models & Capabilities Grok Imagine Image 2.0
Does Grok Image 2.0 Text to Image accept reference images?

No. This route is text-to-image only.

Models & Capabilities Grok Imagine Image 2.0
Which output settings are supported?

The page exposes 1K, 2K and auto, 1:1, 16:9, 9:16, 4:3, 3:4, 2:1, 1:2; use the Playground contract for the final available set.

Models & Capabilities Grok Imagine Image 2.0
How is Grok Image 2.0 Text to Image billed?

Billing is per successful generated image. Resolution or quality differences, when present, are reflected by the live page price.

Models & Capabilities Grok Imagine Image 2.0
How do I call the Grok Image 2.0 Text to Image API?

POST an asynchronous task to /v1/tasks with model set to grok-imagine-image-2.0/text-to-image and put generation fields inside input.

Models & Capabilities Grok Imagine Image 2.0
What is Grok Image 2.0 Image to Image best for?

Edit one public reference image for recoloring, restyling, scene replacement, and detail refinement while following its framing.

Models & Capabilities Grok Imagine Image 2.0
Does Grok Image 2.0 Image to Image accept reference images?

Yes. This route requires reference input and accepts up to 1 image.

Models & Capabilities Grok Imagine Image 2.0
How is Grok Image 2.0 Image to Image billed?

Billing is per successful generated image. Resolution or quality differences, when present, are reflected by the live page price.

Models & Capabilities Grok Imagine Image 2.0
How do I call the Grok Image 2.0 Image to Image API?

POST an asynchronous task to /v1/tasks with model set to grok-imagine-image-2.0/image-to-image and put generation fields inside input.

Models & Capabilities Grok Imagine Image 2.0
What is Qwen Image 2.0 best for?

A cost-efficient Chinese-friendly model with negative prompts, prompt extension, optional seed, and five 2K pixel-size presets.

Models & Capabilities Qwen Image 2.0
Does Qwen Image 2.0 accept reference images?

No. This route is text-to-image only.

Models & Capabilities Qwen Image 2.0
Which output settings are supported?

The page exposes 2688×1536, 2368×1728, 2048×2048, 1728×2368, 1536×2688 and ; use the Playground contract for the final available set.

Models & Capabilities Qwen Image 2.0
How is Qwen Image 2.0 billed?

Billing is per successful generated image. Resolution or quality differences, when present, are reflected by the live page price.

Models & Capabilities Qwen Image 2.0
How do I call the Qwen Image 2.0 API?

POST an asynchronous task to /v1/tasks with model set to qwen-image-2.0 and put generation fields inside input.

Models & Capabilities Qwen Image 2.0
What is Nano Banana best for?

A general-purpose image generator for illustration, photography, product concepts, and social visuals.

Models & Capabilities Nano Banana
Does Nano Banana accept reference images?

No. This route is text-to-image only.

Models & Capabilities Nano Banana
Which output settings are supported?

The page exposes 1K and auto, 1:1, 16:9, 9:16, 3:4, 4:3, 3:2, 2:3, 5:4, 4:5, 21:9; use the Playground contract for the final available set.

Models & Capabilities Nano Banana
How is Nano Banana billed?

Billing is per successful generated image. Resolution or quality differences, when present, are reflected by the live page price.

Models & Capabilities Nano Banana
How do I call the Nano Banana API?

POST an asynchronous task to /v1/tasks with model set to nano-banana and put generation fields inside input.

Models & Capabilities Nano Banana
What is Nano Banana 2 best for?

Create or remix images with optional references, broad aspect ratios, and higher-resolution output.

Models & Capabilities Nano Banana 2
Does Nano Banana 2 accept reference images?

Yes, up to 14 optional references. Omit them for text-only generation.

Models & Capabilities Nano Banana 2
Which output settings are supported?

The page exposes 1K, 2K, 4K and 1:1, 1:4, 1:8, 2:3, 3:2, 3:4, 4:1, 4:3, 4:5, 5:4, 8:1, 9:16, 16:9, 21:9, auto; use the Playground contract for the final available set.

Models & Capabilities Nano Banana 2
How is Nano Banana 2 billed?

Billing is per successful generated image. Resolution or quality differences, when present, are reflected by the live page price.

Models & Capabilities Nano Banana 2
How do I call the Nano Banana 2 API?

POST an asynchronous task to /v1/tasks with model set to nano-banana-2 and put generation fields inside input.

Models & Capabilities Nano Banana 2
What is GPT Image 2 Ext Text to Image best for?

A compatibility route for GPT Image 2 generation with quality and resolution controls; use the primary GPT Image 2 page for the canonical product overview.

Models & Capabilities GPT Image 2
Does GPT Image 2 Ext Text to Image accept reference images?

No. This route is text-to-image only.

Models & Capabilities GPT Image 2
How is GPT Image 2 Ext Text to Image billed?

Billing is per successful generated image. Resolution or quality differences, when present, are reflected by the live page price.

Models & Capabilities GPT Image 2
How do I call the GPT Image 2 Ext Text to Image API?

POST an asynchronous task to /v1/tasks with model set to gpt-image-2/text-to-image@ext and put generation fields inside input.

Models & Capabilities GPT Image 2
What is GPT Image 2 Beta best for?

A legacy beta route retained for compatibility; new integrations should use the primary GPT Image 2 model ID.

Models & Capabilities GPT Image 2
Does GPT Image 2 Beta accept reference images?

No. This route is text-to-image only.

Models & Capabilities GPT Image 2
How is GPT Image 2 Beta billed?

Billing is per successful generated image. Resolution or quality differences, when present, are reflected by the live page price.

Models & Capabilities GPT Image 2
How do I call the GPT Image 2 Beta API?

POST an asynchronous task to /v1/tasks with model set to gpt-image-2/text-to-image@beta and put generation fields inside input.

Models & Capabilities GPT Image 2
What is GPT Image 2 Ext Image to Image best for?

A compatibility editing route with multiple reference images, quality, resolution, and aspect-ratio controls.

Models & Capabilities GPT Image 2
Does GPT Image 2 Ext Image to Image accept reference images?

Yes. This route requires reference input and accepts up to 6 images.

Models & Capabilities GPT Image 2
How is GPT Image 2 Ext Image to Image billed?

Billing is per successful generated image. Resolution or quality differences, when present, are reflected by the live page price.

Models & Capabilities GPT Image 2
How do I call the GPT Image 2 Ext Image to Image API?

POST an asynchronous task to /v1/tasks with model set to gpt-image-2/image-to-image@ext and put generation fields inside input.

Models & Capabilities GPT Image 2
What is Ideogram V4 best for?

Logos, posters, wordmarks, packaging, and signage that require accurate Latin lettering.

Models & Capabilities Ideogram V4
How should I choose a speed tier?

TURBO for quick drafts, BALANCED for speed-quality balance, and QUALITY for final detail and complex signage.

Models & Capabilities Ideogram V4
How do the pricing tiers differ?

TURBO, BALANCED, and QUALITY use separate per-image tiers. See the live page price for the current amounts.

Models & Capabilities Ideogram V4
How do I improve text accuracy?

Spell out the exact text and define type style, hierarchy, placement, and background contrast.

Models & Capabilities Ideogram V4
Does it accept reference images?

No. The current Ideogram V4 page is text-to-image only.

Models & Capabilities Ideogram V4
What is Qwen Image 2.0 Pro strongest at?

Chinese typography, portraits, product visuals, complex concepts, and five 2K-class output presets.

Models & Capabilities
What does prompt extension do?

prompt_extend is on by default and enriches visual detail with a small latency cost. Turn it off when strict wording matters more.

Models & Capabilities
How do I exclude unwanted elements?

Use negative_prompt and keep the positive prompt explicit about subject, text, and composition.

Models & Capabilities
Does it accept reference images?

No. The current Pro text-to-image route is text-only.

Models & Capabilities
Does size change the price?

All five size presets currently use the same per-image tier. See the live page price for the current amount.

Models & Capabilities
What is Seedream 4.5 Text to Image best for?

Chinese-aesthetic scenes, food photography, sci-fi, landscapes, and general creation that needs 2K or 4K output.

Models & Capabilities Seedream 4.5
Do 2K and 4K cost differently?

Both resolutions currently use the same per-image tier. See the live page price for the current amount.

Models & Capabilities Seedream 4.5
Why is aspect ratio required?

The current contract requires explicit aspect_ratio and resolution so output sizing does not rely on an implicit default.

Models & Capabilities Seedream 4.5
Can I upload references?

Not on this route. Use seedream-4.5/image-to-image.

Models & Capabilities Seedream 4.5
How long can the prompt be?

Four to 3,000 characters, in Chinese or English.

Models & Capabilities Seedream 4.5
Which unusual ratios does Wan 2.7 Image support?

Alongside common ratios, it supports 21:9, 8:1, and 1:8 for extreme wide or tall output.

Models & Capabilities Wan 2.7 Image
What does Thinking Mode do?

It adds latency to improve reasoning and composition for complex prompts. Leave it off for fast drafts.

Models & Capabilities Wan 2.7 Image
How should I choose 1K, 2K, or 4K?

Validate direction at 1K or 2K, then move to 4K for larger final output. Current page pricing is flat per image.

Models & Capabilities Wan 2.7 Image
Does it accept references?

No. This model ID is text-to-image only.

Models & Capabilities Wan 2.7 Image
How do I reproduce a result?

Keep seed, prompt, ratio, resolution, and Thinking Mode fixed.

Models & Capabilities Wan 2.7 Image
What is Z-Image best for?

Low-cost Chinese-prompt generation, realistic portraits, signage, urban nights, and food assets.

Models & Capabilities Z-Image
Which aspect ratios are supported?

1:1, 4:3, 3:4, 16:9, and 9:16.

Models & Capabilities Z-Image
Can I control resolution or format?

The current public contract exposes only prompt and aspect_ratio. Do not add undeclared resolution or format fields.

Models & Capabilities Z-Image
Does it accept reference images?

No. The current Z-Image model ID is text-to-image only.

Models & Capabilities Z-Image
How do I call the API?

POST /v1/tasks with model: z-image and input.prompt. aspect_ratio is optional.

Models & Capabilities Z-Image
What type of video model is HappyHorse 1.0?

It is a pure text-to-video route and does not require an opening image.

Models & Capabilities HappyHorse 1.0
Which output settings does HappyHorse 1.0 support?

3-15 seconds, 720p, 1080p, and 16:9, 9:16, 1:1, 4:3, 3:4.

Models & Capabilities HappyHorse 1.0
Can it generate sound or use audio references?

No media input; audio behavior follows the model contract

Models & Capabilities HappyHorse 1.0
How is HappyHorse 1.0 billed?

Billing follows output duration and the live tier. Resolution, mode, or audio differences are reflected by the page price and task record.

Models & Capabilities HappyHorse 1.0
How do I call the HappyHorse 1.0 API?

POST an asynchronous task to /v1/tasks with model set to happyhorse-1.0 and place parameters inside input.

Models & Capabilities HappyHorse 1.0
What type of video model is Kling 3.0 Text-to-Video?

It is a pure text-to-video route and does not require an opening image.

Models & Capabilities Kling 3.0
Which output settings does Kling 3.0 Text-to-Video support?

3-15 seconds, 720p, 1080p, 4K, and 16:9, 9:16, 1:1.

Models & Capabilities Kling 3.0
Can it generate sound or use audio references?

Optional generated sound switch

Models & Capabilities Kling 3.0
How is Kling 3.0 Text-to-Video billed?

Billing follows output duration and the live tier. Resolution, mode, or audio differences are reflected by the page price and task record.

Models & Capabilities Kling 3.0
How do I call the Kling 3.0 Text-to-Video API?

POST an asynchronous task to /v1/tasks with model set to kling-3.0/text-to-video and place parameters inside input.

Models & Capabilities Kling 3.0
What type of video model is Veo 3.1 Fast Text to Video?

It is a pure text-to-video route and does not require an opening image.

Models & Capabilities Veo 3.1 Fast
Which output settings does Veo 3.1 Fast Text to Video support?

4-8 seconds, 720p, 1080p, 4k, and 16:9, 9:16.

Models & Capabilities Veo 3.1 Fast
Can it generate sound or use audio references?

Generated audio enabled by default

Models & Capabilities Veo 3.1 Fast
How is Veo 3.1 Fast Text to Video billed?

Billing follows output duration and the live tier. Resolution, mode, or audio differences are reflected by the page price and task record.

Models & Capabilities Veo 3.1 Fast
How do I call the Veo 3.1 Fast Text to Video API?

POST an asynchronous task to /v1/tasks with model set to veo-3.1-fast/text-to-video and place parameters inside input.

Models & Capabilities Veo 3.1 Fast
What type of video model is Veo 3.1 Text to Video?

It is a pure text-to-video route and does not require an opening image.

Models & Capabilities Veo 3.1
Which output settings does Veo 3.1 Text to Video support?

4-8 seconds, 720p, 1080p, 4k, and 16:9, 9:16.

Models & Capabilities Veo 3.1
Can it generate sound or use audio references?

Generated audio enabled by default

Models & Capabilities Veo 3.1
How is Veo 3.1 Text to Video billed?

Billing follows output duration and the live tier. Resolution, mode, or audio differences are reflected by the page price and task record.

Models & Capabilities Veo 3.1
How do I call the Veo 3.1 Text to Video API?

POST an asynchronous task to /v1/tasks with model set to veo-3.1/text-to-video and place parameters inside input.

Models & Capabilities Veo 3.1
What type of video model is Seedance 2.0?

It is prompt-led and can also use endpoint frames or image, video, and audio references according to the active contract.

Models & Capabilities Seedance 2.0
Which output settings does Seedance 2.0 support?

4-15 seconds, 480p, 720p, 1080p, 4k, and 1:1, 4:3, 3:4, 16:9, 9:16, 21:9, adaptive.

Models & Capabilities Seedance 2.0
Can it generate sound or use audio references?

Optional image, video, and audio references; generated audio

Models & Capabilities Seedance 2.0
How is Seedance 2.0 billed?

Billing follows output duration and the live tier. Resolution, mode, or audio differences are reflected by the page price and task record.

Models & Capabilities Seedance 2.0
How do I call the Seedance 2.0 API?

POST an asynchronous task to /v1/tasks with model set to seedance-2.0 and place parameters inside input.

Models & Capabilities Seedance 2.0
What type of video model is Seedance 2.0 Fast?

It is prompt-led and can also use endpoint frames or image, video, and audio references according to the active contract.

Models & Capabilities
Which output settings does Seedance 2.0 Fast support?

4-15 seconds, 480p, 720p, and 1:1, 4:3, 3:4, 16:9, 9:16, 21:9, adaptive.

Models & Capabilities
Can it generate sound or use audio references?

Optional image, video, and audio references; generated audio

Models & Capabilities
How is Seedance 2.0 Fast billed?

Billing follows output duration and the live tier. Resolution, mode, or audio differences are reflected by the page price and task record.

Models & Capabilities
How do I call the Seedance 2.0 Fast API?

POST an asynchronous task to /v1/tasks with model set to seedance-2.0-fast and place parameters inside input.

Models & Capabilities
What type of video model is Seedance 2.0 Ext?

It is prompt-led and can also use endpoint frames or image, video, and audio references according to the active contract.

Models & Capabilities Seedance 2.0
Which output settings does Seedance 2.0 Ext support?

4-15 seconds, 720p, 1080p, 4k, and 1:1, 4:3, 3:4, 16:9, 9:16, 21:9, adaptive.

Models & Capabilities Seedance 2.0
How is Seedance 2.0 Ext billed?

Billing follows output duration and the live tier. Resolution, mode, or audio differences are reflected by the page price and task record.

Models & Capabilities Seedance 2.0
How do I call the Seedance 2.0 Ext API?

POST an asynchronous task to /v1/tasks with model set to seedance-2.0@ext and place parameters inside input.

Models & Capabilities Seedance 2.0
What type of video model is Seedance 2.0 Mini?

It is prompt-led and can also use endpoint frames or image, video, and audio references according to the active contract.

Models & Capabilities
Which output settings does Seedance 2.0 Mini support?

4-15 seconds, 480p, 720p, and 1:1, 4:3, 3:4, 16:9, 9:16, 21:9, adaptive.

Models & Capabilities
Can it generate sound or use audio references?

Optional image, video, and audio references; generated audio

Models & Capabilities
How is Seedance 2.0 Mini billed?

Billing follows output duration and the live tier. Resolution, mode, or audio differences are reflected by the page price and task record.

Models & Capabilities
How do I call the Seedance 2.0 Mini API?

POST an asynchronous task to /v1/tasks with model set to seedance-2.0-mini and place parameters inside input.

Models & Capabilities
What type of video model is HappyHorse 1.1 Text to Video?

It is a pure text-to-video route and does not require an opening image.

Models & Capabilities HappyHorse 1.1
Which output settings does HappyHorse 1.1 Text to Video support?

3-15 seconds, 720p, 1080p, and 16:9, 9:16, 3:4, 4:3, 4:5, 5:4, 1:1, 9:21, 21:9.

Models & Capabilities HappyHorse 1.1
Can it generate sound or use audio references?

No media input; audio behavior follows the model contract

Models & Capabilities HappyHorse 1.1
How is HappyHorse 1.1 Text to Video billed?

Billing follows output duration and the live tier. Resolution, mode, or audio differences are reflected by the page price and task record.

Models & Capabilities HappyHorse 1.1
How do I call the HappyHorse 1.1 Text to Video API?

POST an asynchronous task to /v1/tasks with model set to happyhorse-1.1/text-to-video and place parameters inside input.

Models & Capabilities HappyHorse 1.1
What type of video model is MiniMax H3?

It is prompt-led and can also use endpoint frames or image, video, and audio references according to the active contract.

Models & Capabilities MiniMax H3
Which output settings does MiniMax H3 support?

4-15 seconds, 2K, and 21:9, 16:9, 4:3, 1:1, 3:4, 9:16, adaptive.

Models & Capabilities MiniMax H3
Can it generate sound or use audio references?

Optional image, video, and audio references; generated audio

Models & Capabilities MiniMax H3
How is MiniMax H3 billed?

Billing follows output duration and the live tier. Resolution, mode, or audio differences are reflected by the page price and task record.

Models & Capabilities MiniMax H3
How do I call the MiniMax H3 API?

POST an asynchronous task to /v1/tasks with model set to minimax-h3 and place parameters inside input.

Models & Capabilities MiniMax H3
What type of video model is Wan 3.0 Video?

It is prompt-led and can also use endpoint frames or image, video, and audio references according to the active contract.

Models & Capabilities Wan 3.0 Video
Which output settings does Wan 3.0 Video support?

2-30 seconds, 480P, 720P, 1080P, and adaptive, 16:9, 4:3, 1:1, 3:4, 9:16.

Models & Capabilities Wan 3.0 Video
Can it generate sound or use audio references?

Optional image, video, audio, file and link references; synchronized audio

Models & Capabilities Wan 3.0 Video
How is Wan 3.0 Video billed?

Billing follows output duration and the live tier. Resolution, mode, or audio differences are reflected by the page price and task record.

Models & Capabilities Wan 3.0 Video
How do I call the Wan 3.0 Video API?

POST an asynchronous task to /v1/tasks with model set to wan3.0-video and place parameters inside input.

Models & Capabilities Wan 3.0 Video
What inputs does Wan 3.0 Prime accept?

Only the prompt is required. Optional inputs are first and last frames, up to 10 reference images, up to 5 reference audio clips, and one reference document or one reference webpage. Reference video input is not available yet.

Models & Capabilities
Can I combine first/last frames with references?

No. First/last-frame mode is mutually exclusive with the reference image, audio, file and webpage arrays, so pick one control mode per task; a reference file and a reference webpage are also either-or.

Models & Capabilities
What are the reference audio requirements?

WAV or MP3, 1 to 15 seconds each, up to five clips, used together with reference images. Refer to them in the prompt as Audio1, Audio2 in order and say whether each is narration, dialogue or music.

Models & Capabilities
Which lengths, resolutions and aspect ratios are supported?

Length is any whole number of seconds from 2 to 30; resolution is 480P, 720P or 1080P; aspect ratio is adaptive, 16:9, 4:3, 1:1, 3:4 or 9:16, with adaptive following the input media and prompt.

Models & Capabilities
Does the video include sound?

By default it generates a synchronized audio track, and you can describe the sound in the prompt (ambience, narration and so on); set audio to false for picture only.

Models & Capabilities
How is Wan 3.0 Prime different from Wan 3.0 Video?

Prime is the fast Wan 3.0 route. Both share the same prompt, first/last-frame and reference image, audio, file and webpage usage, so you switch by changing the model ID; Prime does not accept reference video input yet, and the two routes are priced differently.

Models & Capabilities
How do I reproduce a result?

Pass a fixed seed (0 to 2147483647) and keep the prompt and other settings unchanged to get highly consistent output, which makes it easy to compare after small tweaks.

Models & Capabilities
How is Wan 3.0 Prime billed?

By output video length and resolution tier, and failed tasks are refunded in full; see the live price on this page for the current rate.

Models & Capabilities
Does Seedance 2.5 text-to-video need an image?

No. seedance-2.5/text-to-video accepts a text prompt only. Use the separate Image-to-Video or Reference-to-Video route for frames or reference media.

Models & Capabilities Seedance 2.5
How long can a Seedance 2.5 video be?

The current contract supports 4–30 seconds at 720p or 1080p with seven aspect-ratio options including adaptive.

Models & Capabilities Seedance 2.5
Does Seedance 2.5 generate audio?

generate_audio is enabled by default for synchronized sound and can be turned off. Review volume, dialogue, and rights before production use.

Models & Capabilities Seedance 2.5
How is Seedance 2.5 priced?

It is billed per output second, with separate 720p and 1080p tiers.

Models & Capabilities Seedance 2.5
How do I call the Seedance 2.5 API?

POST an asynchronous task to /v1/tasks with model set to seedance-2.5/text-to-video and pass prompt, duration, resolution, and aspect_ratio inside input.

Models & Capabilities Seedance 2.5
Does Grok Imagine text-to-video require an image?

No. grok-imagine/text-to-video accepts text only; use its image-to-video route for a source frame.

Models & Capabilities Grok Imagine
Which formats does Grok Imagine support?

It supports 6–30 seconds at 480p or 720p in 2:3, 3:2, 1:1, 16:9, or 9:16.

Models & Capabilities Grok Imagine
What are normal and fun modes?

normal targets general motion and steadier shots, while fun is for more energetic creative motion. Start with normal to validate subject and composition.

Models & Capabilities Grok Imagine
How is Grok Imagine priced?

Billing is per output second, with separate 480p and 720p rates.

Models & Capabilities Grok Imagine
How do I call the Grok Imagine API?

POST to /v1/tasks with model set to grok-imagine/text-to-video and place the generation fields inside input.

Models & Capabilities Grok Imagine
Does Hailuo 2.3 text-to-video need an image?

No. This route accepts a text prompt only. Use a separate Hailuo image-to-video model for a source frame.

Models & Capabilities Hailuo 2.3
Which durations does Hailuo 2.3 support?

The standard route offers fixed 6- and 10-second output tiers and does not expose resolution or aspect-ratio fields.

Models & Capabilities Hailuo 2.3
What does prompt_optimizer do?

It is enabled by default to expand and organize the prompt. Turn it off when the original instruction must be preserved exactly.

Models & Capabilities Hailuo 2.3
How is Hailuo 2.3 priced?

It uses fixed per-video prices for the 6- and 10-second tiers.

Models & Capabilities Hailuo 2.3
How do I call the Hailuo 2.3 API?

POST an asynchronous task to /v1/tasks with model set to hailuo-2.3/text-to-video and pass prompt, duration, and prompt_optimizer inside input.

Models & Capabilities Hailuo 2.3
Does Wan 2.7 text-to-video need an image?

No. wan2.7-video/text-to-video requires a prompt only. Use Wan 2.7 Image-to-Video for first or last frames.

Models & Capabilities Wan 2.7 Video
Which output formats does Wan 2.7 support?

It supports 2–15 seconds at 720P or 1080P in 16:9, 9:16, 1:1, 4:3, or 3:4.

Models & Capabilities Wan 2.7 Video
What is audio_url in Wan 2.7?

audio_url is an optional WAV or MP3 reference. It does not turn this route into image-to-video or replace the text prompt.

Models & Capabilities Wan 2.7 Video
How is Wan 2.7 priced?

It is billed per output second with separate 720P and 1080P rates.

Models & Capabilities Wan 2.7 Video
How do I call the Wan 2.7 API?

POST an asynchronous task to /v1/tasks with model set to wan2.7-video/text-to-video and pass prompt, resolution, ratio, and duration inside input.

Models & Capabilities Wan 2.7 Video
How many input images does Veo 3.1 image-to-video accept?

The current endpoint accepts one publicly reachable image URL per request. The field is image_url, not image_urls. Use an image of at least 720p with a clear subject, composition, and motion headroom; localhost addresses and temporary links that are not publicly reachable cannot be used.

Models & Capabilities Veo 3.1
How do I control landscape or vertical output in Veo 3.1 image-to-video?

The output inherits the input image frame, so no separate aspect_ratio field is required. Prepare a 9:16 source for TikTok, Reels, or Shorts, and a 16:9 source for landscape web or advertising placements. Other shapes may be cropped to fit.

Models & Capabilities Veo 3.1
Which durations and resolutions does Veo 3.1 support?

The current HiAPI request contract supports 4, 6, or 8 seconds at 720p, 1080p, or 4K. Duration and resolution affect total cost. Validate motion with a shorter clip before committing to the final resolution.

Models & Capabilities Veo 3.1
Does Veo 3.1 generate native audio, and how do I direct it?

Yes. Write ambience, foley, and musical rhythm into the prompt in event order—for example, request a low impact sound when a product lands. Review volume, dialogue, brand safety, and platform copyright requirements before publishing commercial work.

Models & Capabilities Veo 3.1
How is Veo 3.1 image-to-video priced?

Billing depends on generated duration and resolution. Audio-enabled 720p/1080p uses the base per-second tier, while 4K uses a higher tier. See the live model price and task record for the final charge.

Models & Capabilities Veo 3.1
When should I use Veo 3.1 instead of Veo 3.1 Fast?

Both tiers use the same one-image input and 4/6/8-second resolution structure. Use Fast to test motion, camera, and audio timing, then use standard Veo 3.1 for the final render. A common workflow is a 720p Fast draft followed by a 1080p or 4K standard render.

Models & Capabilities Veo 3.1
How do I call the Veo 3.1 image-to-video API?

Send a POST request to /v1/tasks, set model to veo-3.1/image-to-video, and pass prompt, image_url, resolution, and duration in input. The create request returns a taskId while video generation continues asynchronously. See the API docs for retrieval, callbacks, errors, and the full parameter contract.

Models & Capabilities Veo 3.1
What is the difference between Vidu Q3 Mix and Vidu Q3 Ad?

They share one parameter set — 1-7 reference images, the same 10 sizes, the same 720P/1080P tiers — and the request shape is identical. They differ in length range and intent: Mix is the general-purpose route at 1-16 seconds, while Ad targets advertising at 3-15 seconds and applies marketing-oriented shot cutting, camera work and sound handling. Reach for Ad on product ads and Mix on everything else.

Models & Capabilities
Why is size limited to 10 enumerated values instead of a free resolution?

size is written as a width*height pixel value, but the API accepts only ten fixed values: five frames (16:9, 4:3, 1:1, 3:4, 9:16) at each of the 720P and 1080P tiers. That settles the frame and the resolution tier in a single field with no separate ratio parameter, at the cost of not being able to request an in-between size. At the 1080P tier the 1:1 value is 1808*1808 rather than 1920*1920.

Models & Capabilities
What is the longest clip it can generate?

Mix accepts 1-16 whole seconds with a default of 5. Sixteen seconds is longer than the 5-10 seconds most reference-to-video endpoints allow, which helps when the action needs room to build. Ad uses a different range: 3-15 seconds.

Models & Capabilities
How many reference images can I send, and how do I tell the model who is who?

reference_image_urls takes 1-7 images that must be public HTTP(S) URLs in JPEG, JPG, PNG or WEBP, up to 50MB each and between 1:4 and 4:1 in ratio. Reference them by array order in the prompt, for example “the character from reference image 1 walks into the scene from reference image 2”; the numbers map one to one onto the array.

Models & Capabilities
How is it billed?

Per actual output second, with the resolution tier setting the rate — 720P and 1080P are priced differently. Current rates live on the pricing page, and your HiAPI task history holds the per-task breakdown.

Models & Capabilities
Do audio and watermark affect the price?

No. The rate depends only on the resolution tier and the actual output seconds, and neither switch is priced. audio is off by default for a silent clip; turn it on and the model generates background music and effects that suit the picture. watermark is also off by default; turn it on and “AI generated” appears in the bottom right corner.

Models & Capabilities
How do I choose between this and Vidu Q4 Preview reference-to-video?

It comes down to inputs. Use Q4 Preview when you need voice references so a character speaks in a fixed voice, or when you need more than seven reference images (it takes 1-15 images plus up to 3 MP3 clips). Use Mix when image references are enough, when you want native portrait or square output, and when you would rather pick the frame and the resolution tier in one field — Q4 Preview splits those into resolution and aspect_ratio.

Models & Capabilities
Do the reference images show up as the opening frame?

No. Reference-to-video treats the images as evidence of what the subject looks like and renders a new shot; the references are not used as the first frame. If a reference carries a full background the opening may stay close to it, so describe the environment clearly when you want an entirely new setting. Use the image-to-video route when the first frame must be an exact image.

Models & Capabilities
How long can the prompt be?

Up to 5000 characters in Chinese or English; anything beyond that is truncated. Saying what the subject does, where it appears and how the camera moves gets further than stacking adjectives.

Models & Capabilities
Does fixing the seed make results reproducible?

The seed range is 0-2147483647, and fixing it makes runs from the same inputs noticeably closer. The model is still probabilistic, and the documentation states plainly that an identical seed does not guarantee an identical result. It is more useful for comparing small prompt changes than for exact reproduction.

Models & Capabilities
What is the difference between Vidu Q3 Ad and Vidu Q3 Mix?

The request shape is identical — the same 1-7 reference images, the same 10 sizes, the same 720P/1080P tiers — and they differ in length range and intent. Ad targets advertising at 3-15 seconds and applies marketing-grade shot cutting, camera work and sound; Mix is general purpose at 1-16 seconds with a more even balance of image quality and motion. Reach for Ad on product ads and Mix on narrative, IP or general footage.

Models & Capabilities
Why does Ad use a different length range from Mix?

They are two separate parameter validations: Ad accepts duration as an integer in [3, 15] and Mix in [1, 16], both defaulting to 5 seconds. So Ad cannot produce a 1-2 second fragment, and does not need the long single-take build-up Mix allows.

Models & Capabilities
The output came out at a different length than the duration I set. Which one am I billed for?

The actual output. The documentation states plainly that when the number of shots is clearly above or below duration, the model may adjust the length to keep the story whole — so the clip can run longer or shorter than requested, and billing follows the actual output. Reserve funds using the value you set, but reconcile against the actual seconds on the invoice.

Models & Capabilities
Why is size limited to 10 enumerated values?

size is written as width*height and the API accepts ten fixed values: five frames (16:9, 4:3, 1:1, 3:4, 9:16) at each of the 720P and 1080P tiers. That settles the frame and the resolution tier in one field with no separate ratio parameter, at the cost of no in-between sizes. Note the 1:1 value at 1080P is 1808*1808.

Models & Capabilities
How many reference images can I send, and what are the limits?

reference_image_urls takes 1-7 images that must be public HTTP(S) URLs in JPEG, JPG, PNG or WEBP, up to 50MB each and between 1:4 and 4:1 in ratio. For ads a common pattern is one hero product shot plus one or two in-use shots, referenced by order in the prompt.

Models & Capabilities
How is it billed?

Per actual output second, with the resolution tier setting the rate, and Ad sits slightly below Mix at the same tier. Current rates live on the pricing page, and your HiAPI task history holds the breakdown.

Models & Capabilities
Do audio and watermark affect the price?

No — the rate depends only on the resolution tier and the actual output seconds. audio is off by default; turn it on and the model generates background music or effects that suit the picture, which most ad work wants. watermark is also off by default and stamps “AI generated” bottom right when on. For paid distribution you would normally leave both at their defaults.

Models & Capabilities
How do I choose between this and Vidu Q4 Preview reference-to-video?

Ad's advantages are ad-oriented camera work, shot cutting and sound handling, plus a lower per-second rate. When you need voice references so a model speaks a line, or more than seven reference images, switch to Q4 Preview, which takes 1-15 images plus up to 3 MP3 clips.

Models & Capabilities
How long should the prompt be?

Up to 5000 characters in Chinese or English. Ad clips are short, though, so saying what the product is, what action the viewer sees and how the camera moves is enough — stacked adjectives tend to pull the shot apart. The references already fix the product's look; the prompt carries the motion and the mood.

Models & Capabilities
Does fixing the seed make results reproducible?

The seed range is 0-2147483647, and fixing it makes runs from the same inputs noticeably closer, which helps when comparing prompt variants. The model is still probabilistic and the documentation states that an identical seed does not guarantee an identical result, so treat it as a comparison aid rather than exact reproduction.

Models & Capabilities
Can I set the output aspect ratio?

No. The frame follows the first-frame image completely: landscape in, landscape out; portrait in, portrait out — so prepare a portrait frame when you want 9:16. The first frame's ratio must sit between 1:4 and 4:1; extreme slivers are rejected on submit.

Models & Capabilities Vidu Q4 Preview
What does the audio switch do?

Native audio is on by default and generates dialogue, ambience and effects in sync with the picture. Turn it off for a silent, picture-only clip — useful for looping backgrounds, your own soundtrack or burned-in captions. The switch does not change the price.

Models & Capabilities Vidu Q4 Preview
How do duration and billing relate?

Billing is the submitted whole-second duration times the per-second rate of the resolution you picked; the audio switch and the number of images are not priced. Higher tiers carry a higher rate — 4K is roughly 8.7 times the 540p rate — and the pricing page lists the current amounts.

Models & Capabilities Vidu Q4 Preview
What is each resolution tier good for?

540p for testing prompts and camera moves (and for small looping backgrounds), 720p for most social and web use, 1080p for regular paid placements, 2K for large screens and TV, and 4K for exhibition, cinema-grade picture or shots you plan to crop in post. Draft low, render the keepers high.

Models & Capabilities Vidu Q4 Preview
How is it different from Vidu Q3 Mix?

Vidu Q3 Mix builds a new composition from 1-7 reference images at 720P or 1080P for 1-16 seconds, and those references are not the opening frame. Vidu Q4 Preview Image-to-Video starts on the supplied picture, runs 3-16 seconds and goes up to 4K with native audio. Choose Q4 Preview when the first frame has to be that exact image.

Models & Capabilities Vidu Q4 Preview
How is it different from reference-to-video?

Image-to-video uses your picture as the clip's first frame, so the output opens on that image. Reference-to-video uses pictures as the look of a person, prop or scene and builds a brand-new composition instead; the references never appear as the opening frame.

Models & Capabilities Vidu Q4 Preview
What are the first-frame requirements?

JPG, JPEG, PNG or WebP, up to 50MB, with an aspect ratio between 1:4 and 4:1. Make the first frame's composition and light what you want the finished shot to be, because it fixes the subject, framing and colour of the result.

Models & Capabilities Vidu Q4 Preview
How should I write the prompt?

Describe only what you add: who does what, how the camera travels, and which sounds or line of dialogue belong in the shot. Do not re-describe what is already visible in the image. English prompts are the most stable, Chinese works too, and the limit is 20,000 characters.

Models & Capabilities Vidu Q4 Preview
Is the prompt required?

No. Without a prompt the model infers motion and camera from the first frame; supplying one lets you set the action order, camera path and sound. Either way, billing follows the submitted duration and resolution.

Models & Capabilities Vidu Q4 Preview
What defaults apply when I omit fields?

Duration defaults to 5 seconds, resolution to 720p and audio to on. All three can be overridden per task, and billing follows the submitted whole-second duration and the chosen tier.

Models & Capabilities Vidu Q4 Preview
How many reference images can I send?

1-15. JPG, PNG or WebP, up to 50MB each, at least 128×128 and between 1:4 and 4:1 per image. The prompt refers to them by array order, so the first file is reference image 1 and the second is reference image 2.

Models & Capabilities Vidu Q4 Preview
What are the limits on reference audio?

Up to 3 clips, MP3 only, 3-12 seconds and no more than 50MB each, referenced in order as reference audio 1. Too many clips, the wrong format or the wrong length returns a parameter error at submit time and no task is created.

Models & Capabilities Vidu Q4 Preview
Does a voice reference control the whole soundtrack?

No. It supplies timbre only: when a character speaks, the line in the prompt is delivered in a voice close to the clip, while ambience and effects still follow the picture. Leave the reference audio out and keep audio on when you want ambience without a particular voice.

Models & Capabilities Vidu Q4 Preview
How do I point the prompt at a specific input?

By array order. The first image is reference image 1 and the first clip is reference audio 1; write that number in the prompt. The prompt is required, accepts 1-20,000 characters, and works in Chinese or English.

Models & Capabilities Vidu Q4 Preview
How is the output frame set?

With aspect_ratio: 16:9, 9:16, 1:1, 4:3 or 3:4, defaulting to 16:9. It is independent of the references' own ratios, so square or portrait images can produce a 16:9 master.

Models & Capabilities Vidu Q4 Preview
How long can a clip be, and at what resolutions?

3-16 whole seconds, 5 by default, at 540p, 720p, 1080p, 2K or 4K, defaulting to 720p. Rerun the same prompt at another tier; nothing else has to change.

Models & Capabilities Vidu Q4 Preview
How is it billed?

Billing is the submitted whole-second duration times the per-second price of the chosen resolution. The audio switch and the number of reference images or clips do not change that price. The pricing page carries current amounts, and task history shows each charge.

Models & Capabilities Vidu Q4 Preview
Do the references appear as the opening frame?

No. They describe how the people, props and scenes look, and the model builds a new composition, so the opening is not any one reference. Use image-to-video when the image itself must be the first frame.

Models & Capabilities Vidu Q4 Preview
How is this different from Vidu Q3 Mix?

Q3 Mix takes 1-7 reference images at up to 1080P for 1-16 seconds and accepts no voice reference. Q4 Preview takes 1-15 images plus up to 3 voice clips, up to 4K, for 3-16 seconds. Choose Q4 Preview for a locked voice, more references or 4K.

Models & Capabilities Vidu Q4 Preview
How is this different from Vidu Q3 Ad?

Q3 Ad is built for product ads: 1-7 reference images, 3-15 seconds at 720P or 1080P, with camera work, shot cutting and sound generated for the format. Q4 Preview is built for a recurring cast and a locked voice, takes up to 15 images and 3 clips, and reaches 4K. Use Q3 Ad for product spots, Q4 Preview for series characters or 4K delivery.

Models & Capabilities Vidu Q4 Preview
Can this model generate images?

It accepts text and images and returns text. Use it for image descriptions, document and chart understanding, visual questions, and reasoning over images. Choose an image model to generate or edit images.

Models & Capabilities
Which model ID and endpoint should I use?

Set model to deepseek-v4-flash-vision-exp. Use POST /v1/responses with input, or POST /v1/chat/completions with messages. Their image formats differ, so use the example for your chosen endpoint.

Models & Capabilities
How do I upload images, including multiple images?

The Playground accepts uploaded or pasted JPEG, PNG, WebP and GIF files: up to 4 images per message, 10 MB each. The API accepts public HTTPS image URLs or base64 data URLs. Responses uses input_image; Chat Completions uses image_url content blocks. The stated upload count and size limits apply to the Playground.

Models & Capabilities
How do I select or disable reasoning?

Responses uses reasoning.effort. The page offers none, low, medium and high, defaulting to high. none disables reasoning; other levels request reasoning effort, with latency and token use depending on the task. For Chat Completions, disable reasoning with reasoning.enabled=false.

Models & Capabilities
What are the context and output limits, and why can an answer be cut short?

The context window is 1,048,576 tokens and the output budget limit is 262,144 tokens. The Playground and default examples use 1,024. Reasoning tokens count toward this budget and may consume much of it before the final answer. Raise max_output_tokens or retry without reasoning when output is truncated; combined input and output must still fit the context window.

Models & Capabilities
How are images, reasoning and cached tokens billed?

Image tokens count as input, reasoning tokens count as output, and cached input is billed at the cached-input rate. See the live pricing section for all three rates. Compare usage.input_tokens, usage.output_tokens and input_tokens_details.cached_tokens against your API logs. A cache hit does not make the entire request free.

Models & Capabilities
Do I need to resend conversation history?

The examples use store:false. Resend the required context in input on each turn; a previous text response alone does not retain the conversation for you. With Chat Completions, put the relevant history in messages.

Models & Capabilities
Does it support JSON output and function calls?

Yes. For Responses, use text.format.type=json_object for JSON and tools/tool_choice to define and select functions. Your application executes returned function calls, then supplies a function_call_output with the matching call_id. The model does not execute local code for your application. See the API documentation for complete examples.

Models & Capabilities
How do I detect the end of a streamed response?

Read text from response.output_text.delta and final usage from response.completed. Also handle response.incomplete and response.failed. Do not wait for the Chat Completions [DONE] marker on the Responses endpoint.

Models & Capabilities
Can I reuse an existing HiAPI API key?

Yes, if the key has access to this model and the account has sufficient balance. Set the OpenAI SDK base_url to https://api.hiapi.ai/v1 and use responses.create or chat.completions.create. The web Playground uses your signed-in session, so you do not paste an API key there.

Models & Capabilities
What does Wan 2.6 Flash Image to Video need?

One image and a prompt. The image opens the video; the prompt describes motion, camera and mood.

Models & Capabilities Wan 2.6 Flash
Can I pass several images?

No. image_urls takes exactly one image.

Models & Capabilities Wan 2.6 Flash
What are the image requirements?

JPG, PNG or WebP, at least 256×256 pixels and up to 10MB.

Models & Capabilities Wan 2.6 Flash
Which lengths are supported?

5, 10 or 15 seconds; 5 by default.

Models & Capabilities Wan 2.6 Flash
Which resolutions are supported?

720p and 1080p; 720p by default.

Models & Capabilities Wan 2.6 Flash
How do I set the aspect ratio?

No setting needed — the output follows the input image: landscape in, landscape out; portrait in, portrait out.

Models & Capabilities Wan 2.6 Flash
What does the audio switch do?

With audio on, the model adds ambience and effects synced to the picture. It is off by default, giving a silent clip.

Models & Capabilities Wan 2.6 Flash
Does audio change the price?

Yes. Audio-on clips use the audio per-second rate, silent clips the silent rate, and 1080p costs more than 720p. See the pricing page for current prices.

Models & Capabilities Wan 2.6 Flash
How is it billed?

Per second of video; the per-second rate depends on resolution and whether audio is on.

Models & Capabilities Wan 2.6 Flash
Can I write prompts in Chinese?

Yes. Chinese and English are both supported, up to 1,500 characters.

Models & Capabilities Wan 2.6 Flash
How should I write prompts?

The image already sets subject and scene, so focus on action, camera and mood, e.g. "the fox turns toward the camera, golden leaves drift down, slow push-in".

Models & Capabilities Wan 2.6 Flash
How long are results kept?

Result links expire. Download or copy clips to your own storage if you need to keep them.

Models & Capabilities Wan 2.6 Flash
What is Wan 2.7 Image Standard?

It is the standard tier of the Wan 2.7 image family. One model ID covers prompt-only generation and reference-based editing at 1K or 2K, with up to nine references.

Models & Capabilities
How is it different from Wan 2.7 Image Pro?

The standard tier supports 1K/2K for both text-to-image and reference editing. Pro supports 1K/2K/4K but is text-to-image only. Choose Pro for 4K text-to-image and Standard for generation plus editing up to 2K.

Models & Capabilities
Does it support reference-based editing?

Yes. Pass up to nine public image URLs. Omit them for prompt-only generation; with images, the prompt edits or references them.

Models & Capabilities
Can I set an aspect ratio when images are attached?

No, and you do not need to. The output follows the last reference image, so aspect_ratio is ignored. For prompt-only generation, choose one of eight ratios.

Models & Capabilities
Which resolutions and aspect ratios are supported?

Resolution supports 1K and 2K. Prompt-only generation supports eight ratios: 1:1, 16:9, 4:3, 21:9, 3:4, 9:16, 8:1, and 1:8.

Models & Capabilities
How many images does one request return?

The HiAPI model currently returns one image per request and bills per successfully generated image.

Models & Capabilities
Which reference formats and sizes are supported?

JPEG, PNG, BMP, and WebP up to 20 MB each; each side must be 240–8000 px and the aspect ratio within 1:8–8:1.

Models & Capabilities
Do Chinese prompts and in-image text work?

Chinese and English prompts both work. For in-image text, use a short headline and quote it exactly; Chinese layout depends on character count, typeface, and aspect ratio, so confirm with a 1K proof before final production.

Models & Capabilities
How do I call the Wan 2.7 Image API?

POST an asynchronous task to /v1/tasks with model set to wan2.7-image-standard and prompt inside input; image_urls, aspect_ratio, resolution, and seed are optional. Retrieve the result by task ID.

Models & Capabilities
What inputs does Wan 3.0 Prime accept?

Only the prompt is required. Optional inputs are first and last frames, up to 10 reference images, up to 5 reference audio clips, and one reference document or one reference webpage. Reference video input is not available yet.

Models & Capabilities
Can I combine first/last frames with references?

No. First/last-frame mode is mutually exclusive with the reference image, audio, file and webpage arrays, so pick one control mode per task; a reference file and a reference webpage are also either-or.

Models & Capabilities
What are the reference audio requirements?

WAV or MP3, 1 to 15 seconds each, up to five clips, used together with reference images. Refer to them in the prompt as Audio1, Audio2 in order and say whether each is narration, dialogue or music.

Models & Capabilities
Which lengths, resolutions and aspect ratios are supported?

Length is any whole number of seconds from 2 to 30; resolution is 480P, 720P or 1080P; aspect ratio is adaptive, 16:9, 4:3, 1:1, 3:4 or 9:16, with adaptive following the input media and prompt.

Models & Capabilities
Does the video include sound?

By default it generates a synchronized audio track, and you can describe the sound in the prompt (ambience, narration and so on); set audio to false for picture only.

Models & Capabilities
How is Wan 3.0 Prime different from Wan 3.0 Video?

Prime is the fast Wan 3.0 route. Both share the same prompt, first/last-frame and reference image, audio, file and webpage usage, so you switch by changing the model ID; Prime does not accept reference video input yet, and the two routes are priced differently.

Models & Capabilities
How do I reproduce a result?

Pass a fixed seed (0 to 2147483647) and keep the prompt and other settings unchanged to get highly consistent output, which makes it easy to compare after small tweaks.

Models & Capabilities
How is Wan 3.0 Prime billed?

By output video length and resolution tier, and failed tasks are refunded in full; see the live price on this page for the current rate.

Models & Capabilities

Frequently Asked Questions