Skip to main content
MiniMax-H3 reference-to-video takes a subject/character reference image and keeps that identity consistent across the generated 2K clip, driven by your text prompt. Unlike image-to-video, the reference image is not used as the literal first frame. Clips are billed per second of output.

Endpoint

Async — returns {id, status: "pending"}; poll GET /v1/videos/generations/{id} until success or failed.

Parameters

string
required
Text prompt describing the scene and action for the referenced subject. Max 7000 characters.
string[]
required
A single-element array with the URL of the subject/character reference image whose identity should be preserved across the clip.
"6" | "10"
default:"\"6\""
Length of the generated clip in seconds. Drives the price (billed per second of output).
"2K"
default:"\"2K\""
Output resolution.
"16:9" | "4:3" | "1:1" | "3:4" | "9:16" | "21:9" | "adaptive"
default:"\"16:9\""
Aspect ratio. Defaults to a concrete 16:9; adaptive is also accepted.

Example

Poll until terminal: