Sign in
video-edit

Produce a video from reference media with MiniMax H3

Produce a video from a text prompt and reference media with MiniMax H3.

The prompt addresses the references by position: the first reference image is Image 1, the second Image 2, the first reference video is Video 1, and the first reference audio is Audio 1. Supply reference images either as reference_image_asset_identifiers (images already stored with Ideogram) or as raw reference_images bytes (multipart requests only); supplying both is rejected, and uploaded bytes are used for this request only and are not stored as an asset. Supply reference videos as reference_video_asset_identifiers, which must reference videos generated or uploaded with Ideogram. At most 9 reference images and 3 reference videos are accepted, and reference videos are capped again on clip length. Optional reference_audios accepts MP3/WAV uploads alongside an image or video. Audio and video references must total at most 15 seconds; at most 12 references are accepted across all media.

References are optional. With none, the video is produced from the prompt alone.

Video generation always runs asynchronously: the response returns as soon as the request is accepted and carries only a generation_id. Poll for completion and results with GET /v1/generations/{generation_id} using that id, or supply a webhook_url to have the finished result POSTed to your server instead.

Video links are available for a limited period of time; download the video if you want to keep it.

post/v2/video/generate/minimax-h3-reference-to-video

Query parameters

dry_runboolean

When true, the request is validated and priced but not run: nothing is generated, stored, or billed, and no safety review is performed. The response is a PriceQuote object instead of the usual response for this endpoint. Send exactly the request you would send to generate, so the quote reflects the same options.

Request body

promptstring required

A natural-language prompt describing the video to produce. Reference media is addressed by position, as in "Image 1 walks toward the camera with the motion of Video 1".

reference_imagesstring[]

Images to use as references (max size 50MB each), as raw bytes, in prompt order; only common image formats such as JPEG, PNG, and WEBP are supported. Multipart requests only. Cannot be combined with reference_image_asset_identifiers. The bytes are used for this request only and are not stored as an asset.

reference_audiosstring[]

MP3 or WAV audio references, in prompt order as Audio 1, Audio 2, and Audio 3. Multipart uploads only; up to 15 MB per file and 2–15 seconds each. Audio and video references must total no more than 15 seconds. Requires a reference image or video. At most 12 references across images, videos, and audio. Used only for this request and not saved as assets.

aspect_ratio'21x9' | '16x9' | '4x3' | '1x1' | '3x4' | '9x16'

The aspect ratio of the generated video.

resolution'480P' | '768P' | '2K' | '4K'

The resolution tier of the generated video, spelled the way MiniMax spells it. 480P and 768P are generated natively; 2K and 4K are upscaled from a 768P result. Higher tiers cost more.

durationinteger

The length of the generated video in seconds.

prompt_expansion_mode'disabled' | 'fast' | 'balanced' | 'quality'

How much the model may rewrite the prompt before generating. disabled uses the prompt as written; the other modes trade latency for a richer rewrite.

webhook_urlstring uri

HTTPS URL that Ideogram delivers the generated result to. Ideogram sends a JSON POST to this URL once all images for the request have finished generating. The body mirrors the synchronous generate response: request_id, created, and a data array containing every generated image (url, prompt, resolution, seed, is_image_safe). Each delivery is signed with Ed25519 and verifiable against the public keys at https://api.ideogram.ai/v1/.well-known/jwks.json. Must be HTTPS; private and loopback hosts and the cloud metadata service are rejected.

privateboolean nullable

When true or omitted, the output is kept private to your account. Set to false to publish the output to the public feed. Enterprise accounts always generate privately.

target_collection_idstring

A collection you can write to, by its URL-safe base64 collection id. The output videos are added to it when the request completes.

Example request

{
  "prompt": "Image 1 walks through the snowy forest at dawn.",
  "reference_image_asset_identifiers": [
    {
      "asset_type": "RESPONSE",
      "asset_id": "7uS_VESkRI6O3-sVgHQp_A"
    }
  ],
  "reference_video_asset_identifiers": [
    {
      "asset_type": "RESPONSE",
      "asset_id": "7uS_VESkRI6O3-sVgHQp_A"
    }
  ],
  "duration": 5,
  "webhook_url": "https://api.example.com/webhooks/ideogram"
}

Response

An acknowledgement to poll with GET /v1/generations/{generation_id}.

generation_idstring required

URL-safe base64 ID of the accepted generation. Accepted by the GET /v1/generations/{generation_id} polling endpoint.

createdstring date-time required

The time the request was accepted.

Example response

{
  "generation_id": "generation_id",
  "created": "2000-01-23T04:56:07+00:00"
}

Changes

Changed in 1 of the 33 revisions of this API.1