MiniMax-H3 Video MiniMax-H3 provides three entry points: the official format, the OpenAI-compatible format, and the unified video format. All support text-to-video, image-to-video, and multimodal reference-to-video, as well as regenerating a 768P result into 2K.
Create Video POST /v1/video/generations, create a task with a multimodal content array.
Query Task GET /v1/video/generations/{task_id}, query official format task status and results.
Video Generation POST /v1/videos, create an async video task.
2K Regeneration POST /v1/videos/{video_id}/remix, regenerate a spec-compliant 768P result into 2K.
Task Query GET /v1/videos/{task_id}, query task status and results.
Create Video POST /v1/video/create, create a task using the unified JSON format.
2K Regeneration POST /v1/video/remix, pass the source task ID in the request body.
Query Task GET /v1/video/query?id=, query task status and results.
Supported Capabilities Image-to-video and reference-video/reference-audio generation are mutually exclusive and cannot be mixed in the same request.
Model and Limits Endpoint Mapping Related Pages 