Guide
PixVerse V6 Max Developer Guide: Capabilities, Pricing, and API Integration
A practical route-by-route walkthrough of text-to-video and image-to-video, parameter choices, and per-second cost math
Auf dieser Seite
What PixVerse V6 Max Gives You as a Video Model
PixVerse V6 Max is a video generation model family available through the HiAPI unified API. It returns video as its output type, and that single family is exposed as two distinct routes rather than one endpoint with a mode flag. Each route maps to its own model reference: pixverse-v6-max-text-to-video for prompt-only generation and pixverse-v6-max-image-to-video for animating a supplied first frame.
The split matters because the two routes accept different inputs, enforce different prompt limits, and handle aspect ratio differently. Text-to-video is the open-ended route — you describe a scene and the model builds everything. Image-to-video is the constrained route — you hand over one image and describe how it should move. Both share the same billing unit (per second of output), the same resolution enum, and the same audio toggle, so once you understand one route the other is mostly a matter of swapping required fields.
For most teams the decision is mechanical: if you have a concrete visual you want to animate, use image-to-video; if you need a scene that does not exist yet, use text-to-video. The rest of this guide walks the parameters, the per-second cost math, and the integration details for both. You can browse each route's schema on its model page: text-to-video and image-to-video.
Text-to-Video vs Image-to-Video: Choosing the Right Route
The two routes differ along four practical axes.
Required inputs
Text-to-video requires only prompt. Image-to-video requires both prompt and image_urls, and image_urls accepts exactly one public image URL — it is an array with a minimum and maximum of one item, not a batch field. That single image is used as the first frame of the generated clip.
Prompt length
This is the sharpest difference. Text-to-video allows prompts up to 5000 characters; image-to-video caps at 2048. If your scene description is long and detailed, the text route gives you roughly twice the headroom. On the image route, keep the prompt focused on motion and atmosphere rather than re-describing what is already visible in the frame.
Aspect ratio
Only text-to-video exposes an aspect_ratio parameter, with an enum of 16:9, 4:3, 1:1, 3:4, 9:16, 2:3, 3:2, and 21:9, defaulting to 16:9. Image-to-video has no such parameter: the output follows the aspect ratio of the input image. To control framing on the image route, crop the source image before uploading it.
Audio support
Both routes support generate_audio, and it is off by default on both. When enabled, the model generates background music, sound effects, and dialogue alongside the video. The prompt convention is the same on either route: put spoken lines in quotes inside the prompt so the model knows what should be voiced. Because audio changes the per-second rate, it is a cost decision as much as a creative one — covered in the next sections.
In short: pick text-to-video when you need full scene control, flexible aspect ratios, and long prompts. Pick image-to-video when a specific visual must anchor the first frame and you want the rest of the clip to follow its proportions.
Key API Parameters and What Each One Changes
Most parameters are shared across both routes; a few are route-specific. Here is the full set with defaults and ranges.
prompt (required, both routes)
A string describing the video. Minimum length 1. On text-to-video the maximum is 5000 characters; on image-to-video it is 2048. On the image route, write about motion and atmosphere.
duration (both routes)
An integer in seconds, default 5, minimum 1, maximum 15. This value is the primary billing multiplier, since the unit is per second of output.
resolution (both routes)
A string enum of 360p, 540p, 720p, 1080p, defaulting to 720p. Higher resolutions cost more per second; 720p is the default tier.
generate_audio (both routes)
A boolean, default False. Turning it on generates background music, sound effects, and dialogue. Spoken lines must be quoted inside the prompt. Enabling audio raises the per-second rate at every resolution.
seed (both routes)
An integer from 0 to 2147483647. The same seed and the same prompt reproduce a similar result, which makes it the main lever for repeatable output.
aspect_ratio (text-to-video only)
A string enum of 16:9, 4:3, 1:1, 3:4, 9:16, 2:3, 3:2, 21:9, default 16:9. Ignored or absent on the image route, where the first frame's proportions govern the output.
image_urls (image-to-video only, required)
An array of exactly one public image URL (minItems 1, maxItems 1). Accepted formats are JPEG, PNG, or WebP, up to 20 MB. This image becomes the first frame, and the output inherits its aspect ratio.
A reasonable habit is to fix seed, resolution, and duration first, then iterate on the prompt — that keeps cost per experiment predictable while you tune the creative content.
How Per-Second Pricing Works Across Resolution and Audio
Billing on both routes is measured per second of generated output, and duration is the stated price factor. The effective rate per second depends on two things: the resolution tier and whether audio is enabled. That produces eight tiers, identical across both routes.
The tier table in prose
At 360P with sound off, the rate is the lowest in the family. Moving to 540P with sound off raises it, and 720P with sound off is the default tier rate of $0.0559 per second. Official per-second rates with sound off are $0.0305 (360P), $0.0458 (540P), $0.0559 (720P), and $0.1118 (1080P); with sound on they are $0.0458 (360P), $0.0559 (540P), $0.0762 (720P), and $0.1423 (1080P). 1080P with sound off roughly doubles the 720P silent rate. Turning sound on shifts every resolution up one step in effective cost: 360P with sound on matches the 540P silent rate, 540P with sound on matches the 720P silent rate, and 720P with sound on sits above the 720P silent tier. 1080P with sound on is the most expensive combination in the family.
The practical reading: silent 720p is the baseline you should assume when budgeting, enabling audio or jumping to 1080P are the two fastest ways to raise unit cost, and dropping to 360P or 540P is the fastest way to cut it. Because the tiers are the same on text-to-video and image-to-video, you can compare route economics purely on duration and tier, independent of which route you call.
Current per-second rates for every tier live on the pricing panel; treat those numbers as the source of truth and recompute your budget whenever they change.
Cost Examples for Common Durations
Because the unit is per second, total cost is simply duration multiplied by the tier rate. That makes duration the easiest dial to reason about: halving a clip halves the cost at any tier, and doubling it doubles the cost.
A 5-second clip (the default)
With the default duration of 5 and the default 720p resolution and audio off, you pay the 720P silent rate five times. That is the baseline cost of a single default call, and it is what you should use as your unit of comparison when evaluating other tiers.
Scaling to longer clips
At the same 720P silent tier, a 10-second clip costs twice the 5-second figure and a 15-second clip — the maximum duration — costs three times it. The same linear scaling applies if you move to a higher tier: a 15-second 1080P clip with sound is the maximum single-call cost in the family, since it combines the longest allowed duration with the most expensive per-second tier.
Comparing tiers at fixed duration
Hold duration at 5 seconds and compare: the 360P silent tier is the cheapest, 540P silent sits in the middle, 720P silent is the default, and 1080P silent is roughly double it. Switching audio on at 720P moves you to the 720P sound-on rate, which is higher than the silent 1080P-adjacent step. This is why audio is worth thinking about explicitly — for many clips, silent 1080P and 720P with sound land in similar cost territory, so the choice is creative rather than purely budgetary.
Draft cheap, finalize expensive
A cost-efficient pattern is to iterate at 360P or 540P with audio off while you refine the prompt and seed, then re-run the winning settings once at your target resolution and audio setting. Since seeds are reproducible, the final render uses the same creative decisions you validated in the cheap passes.
Integration Workflow and Reproducibility Tips
The request flow is the same on both routes: build a payload, call the model reference, then retrieve the resulting video. What differs is the required fields and the parameter set.
Build the payload
Start with the required fields. For text-to-video, that means prompt plus your chosen aspect_ratio, resolution, duration, generate_audio, and seed. For image-to-video, include prompt and image_urls (one public URL, JPEG/PNG/WebP, up to 20 MB), then the shared resolution, duration, generate_audio, and seed. Keep the prompt within the route's limit — 5000 characters for text-to-video, 2048 for image-to-video.
Set the cost dials before you iterate
Decide resolution, duration, and generate_audio up front so your per-call cost is stable while you experiment. Duration is bounded to 1–15 seconds, so very long shots are out of scope for a single call — plan sequences as multiple generations.
Use seeds for reproducibility
seed accepts 0 to 2147483647. The same seed with the same prompt produces a similar result. That gives you a repeatable loop: fix the seed, change one prompt variable at a time, and diff the outputs. When you find a good result, record the seed alongside the prompt so you can regenerate it later at a higher resolution or with audio enabled.
Structure prompts for audio
If generate_audio is on, wrap any spoken lines in quotes inside the prompt. On the image route, keep the prompt about motion and atmosphere since the visual content is already fixed by the first frame.
Pick the right route reference
Call the model reference that matches your input: pixverse-v6-max-text-to-video when you have only a prompt, pixverse-v6-max-image-to-video when you have a first-frame image. Both are documented on their model pages — text-to-video and image-to-video — which are the best place to confirm the current parameter schema before wiring up production calls.
Common Pitfalls and Limits to Design Around
Most integration problems come from ignoring a hard limit rather than getting the creative wrong. Here is what to design around.
Prompt length caps differ by route
Text-to-video allows up to 5000 characters; image-to-video caps at 2048. A prompt that works on one route may be rejected or truncated on the other. If you share prompt templates across routes, enforce the shorter limit or branch by route.
One image, not many
image_urls is an array with minItems 1 and maxItems 1. Passing multiple images is not supported — the route builds the clip from a single first frame. If you need multiple shots, issue multiple calls.
Image format and size
Source images must be JPEG, PNG, or WebP and no larger than 20 MB. They must be publicly reachable URLs. Also remember that the output aspect ratio follows the image, so get the crop right before uploading — there is no aspect_ratio override on this route.
Duration range is 1–15 seconds
Both routes accept duration between 1 and 15 seconds, default 5. There is no way to request a longer single clip; long-form video needs to be assembled from multiple generations.
Audio is off unless you ask
generate_audio defaults to False. If your clip needs sound, set it explicitly — and quote spoken lines in the prompt. Enabling it also moves you into a higher per-second tier, so budget for it deliberately.
Resolution enum is fixed
Only 360p, 540p, 720p, and 1080p are valid. Do not assume intermediate or higher tiers exist; validate the value before sending.
Seeds help, they do not guarantee identity
A fixed seed with a fixed prompt reproduces a similar result, not a bit-identical one. Treat seeds as a strong consistency tool for iteration, not as a render cache.
Designing your client around these limits — validating prompt length per route, checking image format and size, and clamping duration and resolution to the enums — avoids most failed calls before they happen. Both routes are documented on the HiAPI model pages, which is the right place to confirm schema before you ship.
Ähnliche Artikel
Alle Artikel ansehen →-
Guide
gpt-6.2-sol Getting Started Guide: Capability Boundaries, API Integration, and Token Pricing Essentials
Integration decisions and practical essentials for engineering teams
-
Guide
gpt-6.2-astra Getting Started Guide: Text Model Capability Boundaries, Integration Essentials, and Cost Control
-
Guide
Tripo H3.1 Getting Started Guide: Choosing Among Three Input Modes, Parameters, and Pricing
A practical guide for developers on 3D generation integration decisions and cost control