MiniMax H3 Max Turbo Guide: Text-to-Video, First/Last-Frame Image-to-Video, Parameters and Pricing

MiniMax H3 Max Turbo is the high-speed tier of H3 Max. HiAPI offers it under two model IDs: minimax-h3-max-turbo/text-to-video (text to video) and minimax-h3-max-turbo/image-to-video (video from a first frame, with an optional last frame). Each clip runs 1–15 seconds at 480P, 768P or 1080P, billed per second. This guide covers which ID to use, how to fill in the parameters and how pricing works.

Which ID to use

• You only have a text description: use text-to-video. The only required field is prompt; set the frame shape with aspect_ratio.
• You already have the shot: use image-to-video with prompt and image_url (the first frame). Add end_image_url as a last frame and the model fills in the motion between the two.
• For image-to-video the output shape follows the first frame, and there is no aspect_ratio field. For a vertical video, pass a vertical image, and keep the first and last frames at the same ratio.

Parameters

• resolution: 480P / 768P / 1080P, default 768P. 1080P is a refinement pass over a 768P render; native generation goes up to 768P.
• duration: 1–15 seconds, default 5, decimals allowed. The output can run up to about 0.7 s longer than requested — ask for 5 s and you get roughly 5.2 s.
• aspect_ratio (text-to-video only): 21:9, 16:9, 4:3, 1:1, 3:4, 9:16, default 16:9. For example, 768P at 3:4 outputs 768×1024 and 768P at 21:9 outputs 1536×672.
• prompt_expansion_mode: prompt rewriting. disabled uses your prompt as is; balanced (default) rewrites it in about a second; quality spends up to about 30 seconds on a richer rewrite.
• target_audio_url: optional. An audio clip of at least 2 seconds and at most 15 MB becomes the soundtrack; anything longer than the video is trimmed. When you use it, set duration to 2 or more.
• seed: optional. Fix it to reproduce a result, keeping every other parameter the same.
• Each request returns one video. Reference images for locking a character, video extension and video editing are not available on these IDs.

Example request

Submit a task:

curl https://api.hiapi.ai/v1/tasks \
  -H "Authorization: Bearer $HIAPI_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax-h3-max-turbo/image-to-video",
    "input": {
      "prompt": "The cyclist slows near the street lamp and turns to look back, rain streaks in warm neon light, slow push-in",
      "image_url": "https://example.com/first-frame.png",
      "resolution": "768P",
      "duration": 5
    }
  }'

You get data.taskId back; poll GET https://api.hiapi.ai/v1/tasks/<taskId>. When status is success, the video URL is in data.output[0].url. A 5-second 768P clip usually finishes in under 20 seconds. Result links expire (see expireAt), so download anything you want to keep.

Pricing

• 480P: $0.025 per second.
• 768P: $0.04 per second.
• 1080P: $0.08 per second.
• Billing uses the duration you request, and both IDs cost the same. For example, 5 seconds at 768P is 5 × 0.04 = $0.20; 10 seconds at 1080P is $0.80.
• You are charged only for tasks that succeed. Failed tasks are refunded in full.

Writing prompts

• For text-to-video, describe subject, setting, action, camera movement, then light and mood.
• For image-to-video the first frame already fixes the composition. Spend fewer words on what is visible and more on motion: who moves, how the camera travels, and when — for example, "slow push-in, the rider turns around at second 3".
• Draft motion and camera work at 480P with short clips, then render the keeper at 768P or 1080P. 1080P costs 3.2 times as much per second as 480P.