Open Weights video generation API

MiniMax H3 SH API

Turn text, frames, and references into video with native audio.

MiniMax H3 SH brings the Open Weights edition to three dedicated models: text-to-video, image-to-video, and reference-to-video. Generate 3–15 second clips in four resolutions through the same task API, with native stereo audio included.

Try playground

3–15s

Output duration

480p–1080p

Resolution

8 credits/s

Starting price

MiniMax

minimax-h3-sh/text-to-videoH3 family sample

3–15s

Output duration

480p–1080p

Resolution

8 credits/s

Starting price

Model capabilities

Built for production API workflows, not one-off demos.

Use MiniMax H3 SH from the same platform surface as the rest of your video and image stack: API keys, credits, logs, webhooks, and docs stay consistent across providers.

Text-to-video

Describe the scene, motion, camera, and soundtrack. Choose one of seven aspect ratios, including 21:9 and 9:21.

First and last frames

Provide one image for the opening frame or two for first and last frames. Image-to-video follows the image aspect ratio; frame images are free.

Multimodal references

Guide a scene with up to 9 images, 3 videos, and 3 audio tracks. Reference videos share a 15-second processing budget; standalone audio is trimmed to 15 seconds per track.

Four output resolutions

Choose 480p, 540p, native 768p, or 1080p. Every mode includes native stereo audio; defaults are 480p and 5 seconds.

API workflow

Submit tasks, track progress, and return generated assets.

Choose a model

Select minimax-h3-sh/text-to-video, minimax-h3-sh/image-to-video, or minimax-h3-sh/reference-to-video. Use image_urls, video_urls, and audio_urls for supported media.

Set the scene and output

Provide a non-empty prompt, choose a duration from 3 through 15 seconds and a resolution, and optionally set a seed. Reference prompts can use <Picture 1>, <Video 1>, and <Audio 1> tags.

Estimate and retrieve

POST /v1/estimate returns the full charge, including reference media. Submit to /v1/videos/generations, then poll the returned task ID or receive a callback and read output.urls.

POST /v1/videos/generations

curl -X POST https://api.aivideoapi.ai/v1/videos/generations \
  -H "Authorization: Bearer sk-your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax-h3-sh/text-to-video",
    "input": {
      "prompt": "A slow tracking shot through a sunlit forest. Audio: birdsong and gentle wind.",
      "resolution": "480p",
      "duration": 5,
      "aspect_ratio": "16:9"
    }
  }'

What teams build

Social video variations

Product image animation

Character and motion references

Music and voice guidance

Storyboard transitions

Cinematic short clips

Frequently asked questions

Answers about the MiniMax H3 SH API.

How does SH differ from the existing MiniMax H3 models?

SH is the Open Weights edition with dedicated model IDs, four resolutions, and its own pricing. Existing minimax-h3, Fast, Max, and Max Turbo models keep their original parameters and prices.

How is reference video duration billed?

The server measures each video, sums the actual durations including fractional seconds, and caps the total at 15 seconds. Two 5.167-second references therefore bill 10.334 input seconds. A 5-second 480p output with those references costs 153.34 credits before account multipliers.

Are longer generated clips billed again?

Output is billed by the requested integer duration. The model's frame grid may produce a slightly longer file, such as about 5.2 seconds for a 5-second request; that does not add a charge.

Can I use audio as the only reference?

Reference-to-video accepts at least one reference image, video, or audio. Each standalone audio reference costs 4 credits; a video's embedded soundtrack does not count as another audio reference.

What happens when a task fails?

A confirmed failed task is refunded once. A successful generation is never refunded because a later delivery or callback attempt needs retrying. Check the task endpoint for the final status.

Pricing and usage

Clear model options with shared credits.

Text / image output

8 / 12 / 16 / 32 credits/s

480p / 540p / 768p / 1080p; frame images are free

Reference output + video input

10 / 15 / 25 / 50 credits/s

Same resolution order; sum actual reference video seconds, including fractions, then cap at 15s

Reference image / audio

4 credits each

Per image or standalone audio track in reference-to-video; native generated audio is included

Start building with MiniMax H3 SH in AI Video API.

Create one API key, use one credit balance, and switch between video and image models without provider-specific plumbing.

Read docs