MiniMax-H3 Video Generation

Relay integration for the MiniMax-H3 (minimax_h3) video generation model: call flow and capability overview.

MiniMax-H3 is a video generation model from MiniMax, exposed through the relay service under the OpenAI Video compatible protocol with a unified model name of minimax_h3.

Integration Info

Item	Value
Base URL	`http://103.36.63.156:3000`
Model Name	`minimax_h3`
Create Task	`POST /v1/videos`
Query Task	`GET /v1/videos/{task_id}`
Download Video	`GET /v1/videos/{task_id}/content`
Authentication	`Authorization: Bearer {api_key}`

Available Capabilities

Capability	Description
Text-to-video	Generate video from a text prompt only
Image-to-video	Pass an image as the first frame
First/last frame	Specify both the first and last frames
Image upload	Supports URL references and multipart upload
Audio reference	Drive or reference with audio
Video reference	Use a video as reference (including its audio track)

Full Call Flow

Call POST /v1/videos to submit a generation task and obtain a task_id.
Poll GET /v1/videos/{task_id} to query the task status.
Once the status becomes completed, download the video via GET /v1/videos/{task_id}/content.

Minimal Working Example

curl -X POST "http://103.36.63.156:3000/v1/videos" \
  -H "Authorization: Bearer sk-xxxxxxxx" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "minimax_h3",
    "prompt": "A small white kitten chasing butterflies on a lawn, bright sunlight, shallow depth of field",
    "size": "1280x720",
    "duration": 6
  }'

Once the response returns a task_id, poll the query endpoint to obtain the result.

Text to Video

Generate video from a text prompt only.

Image to Video

Image URLs, multipart upload, and first/last frames.

Reference Assets

How to use audio and video references.

Query and Download

Task status polling and video download.