Create a video job (asynchronous)

Server URLloading...

POST

/v1/videos

Send

Authorization

Body

Asynchronous. This endpoint submits a job and returns immediately; it does not block until the video is rendered. Poll GET /v1/videos/{id} for status, then fetch the result from GET /v1/videos/{id}/content.

Billing happens when the job completes successfully. Failed jobs are not billed.

Image-to-video. Send input images in the same request. As JSON, each image is an https:// URL or a base64 data URL (data:image/png;base64,...); as multipart/form-data (what the OpenAI SDK sends for input_reference), use the same field names as file parts.

• first_frame_image (aliases input_reference, image): the first frame
• last_frame_image (alias last_frame): the last frame; most models need a first frame with it
• reference_images (array; alias reference_image): reference images
• keyframes (array of { "image": ..., "timestamp_s": 2.5 }): images pinned at a moment inside the clip. In multipart/form-data, send the whole array as one JSON text field.
• reference_videos (array; alias reference_video): reference video clips, MP4 or MOV, up to 50 MB each.
• reference_audios (array): either preset voices { "voice_id": "eve" }, or reference audio files (an https:// URL, a data URL, or { "url": ... }), WAV or MP3, up to 15 MB each. In multipart/form-data, send voices as one JSON text field and audio files as file parts.

Images are PNG, JPEG or WebP, up to 10 MB each, at most 20 images and 80 MB of images per request; all inputs together at most 120 MB. Which inputs a model accepts is listed on its model page: minimax/minimax-hailuo-2.3 takes a first frame only; minimax/minimax-h3 and its variants (-highspeed, -lite, -max) take a first frame or first + last frame, or references (images, videos, audio files; see below); xai/grok-imagine-video-1.5 takes frames, keyframes, reference images and preset voices (see below). On H3, with input images or videos ratio defaults to adaptive (follows the input). An input the model does not support is rejected with a 400, never silently ignored.

MiniMax H3 (minimax/minimax-h3) parameters.

H3 references can be combined with each other, but not with first/last frames:

Across the H3 family a request takes at most 12 input files (images, videos and audio combined).

H3 family. minimax/minimax-h3-highspeed, minimax/minimax-h3-lite and minimax/minimax-h3-max take the same fields, frames and references as H3, with these differences:

Billing is per output second at the rate for the chosen size; 2K costs more per second than 768P. On top of that, reference video is billed per input second at the same rate as the output (each clip rounded up to a whole second), the first 5 reference images are free and each one after that is billed, and reference audio is free. See the model page (and each variant's own page) for the current rates. One video per job (n must be 1).

Not supported on the H3 family through RouteMux: callback_url (poll the job instead). A last frame needs a first frame. Invalid combinations are rejected with 400 MODEL_PARAMETER_UNSUPPORTED before a job is created (and nothing is billed): seconds outside the model's range or not a whole number, an unsupported size or ratio, adaptive without an input image or video, first/last frames combined with references, more than 9 reference images, more than 12 input files, prompt_expansion_mode on a model other than H3 Max, or reference clips outside the limits above. A GIF image, or a video or audio file in another format, returns 415 REQUEST_MEDIA_TYPE_UNSUPPORTED.

xAI Grok Imagine Video 1.5 (xai/grok-imagine-video-1.5) parameters. The output video has sound.

Inputs on Grok Imagine Video 1.5 can be combined in one request:

With reference images or voices the video is capped at 720p; 1080p is rejected with a 400.

Billing is per output second at the rate for the chosen size, plus a charge for each input image (first frame, last frame, keyframe or reference image); voices are free. See the model page for the current rates. One video per job (n must be 1), and a prompt is required with input images too. Video edit / extend is not available (xAI offers it only on the older grok-imagine-video). Invalid values (seconds outside 1–15 or not a whole number, another size or ratio, too many inputs, a keyframe time outside the clip) are rejected with 400 MODEL_PARAMETER_UNSUPPORTED before a job is created, and nothing is billed. Video generation on this model also has a per-account daily allowance: past it, requests return 429 LIMIT_VIDEO_DAILY_CAP until 00:00 UTC.

Response Body

curl -X POST "https://example.com/v1/videos" \  -H "Content-Type: application/json" \  -d '{    "model": "minimax/minimax-h3",    "prompt": "A drone shot over a coastline",    "seconds": 6  }'

{  "model": "string",  "stream": true}