Generate video with MiniMax H3
MiniMax H3 — multimodal 2K video. Text-to-video; image-to-video (start + optional end frame); and reference-to-video mixing up to 5 reference images, up to 3 reference video clips (2-15s each, 15s combined — billed per second of INPUT on top of the output, see POST /v1/cost), and up to 3 reference audio clips (15s combined, free). Reference media can’t be combined with start/end frames. Any duration from 4 to 15 seconds.
Asynchronous. Returns a job id (the video is not in this response). Poll GET /v1/videos/{id} until status is succeeded — the video URL is in that response’s video_url.
Authorizations
API key as a bearer token: Authorization: Bearer bpx_live_xxx
Body
Text description of the video.
Aspect ratio for text-to-video. Ignored (adaptive) when any image or reference input sets the frame.
21:9, 16:9, 9:16, 1:1, 4:3, 3:4 Optional end frame, used together with image (the start frame) to interpolate the video between the two frames. Same input forms as image.
Optional start frame for image-to-video. Provide a public image URL, a data URI, raw base64, or an uploaded-file URL from POST /v1/uploads. Omit for text-to-video. Can't be combined with reference media.
Output video length in seconds (4-15).
4 <= x <= 15Optional reference audio clips (up to 3; each 2-15s, 15s combined; WAV/MP3) that guide sound/voice in reference-to-video mode. Each item is a public audio URL or an uploaded-file URL from POST /v1/uploads. Free — no input billing. Can't be combined with image/end_image.
3Optional reference images (up to 5) that guide identity/style in reference-to-video mode. Each item is a public image URL, a data URI, raw base64, or an uploaded-file URL from POST /v1/uploads. Can't be combined with image/end_image.
5Optional reference video clips (up to 3; each 2-15s, 15s combined; MP4/WebM, ≤30MB each) that guide motion/identity in reference-to-video mode. Each item is a public video URL or an uploaded-file URL from POST /v1/uploads (videos are passed by URL, not inlined). Billed at the model's per-second rate on the INPUT duration in addition to the output — estimate via POST /v1/cost. Can't be combined with image/end_image.
3