MiniMax-H3

Supports text-to-video, image-to-video (first/last frame), and multimodal reference. Resolution 480P / 768P / 1080P / 2K, duration 4–30 seconds.

Action	Method	Path
Create	`POST`	`/v1/videos` (also `/v1/video/generations`)
H3 Context IR	`POST`	`/v1/videos/h3_context_ir`
768P → 2K regen	`POST`	`/v1/videos/regeneration`
Get result	`GET`	`/v1/videos/{task_id}`

Host: https://api.omniall.ai · Auth Authorization: Bearer sk-... · Model: MiniMax-H3

Resolution & duration

Field	Notes
`resolution`	`480P` / `768P` / `1080P` / `2K`
`duration` / `seconds`	4–30 seconds (integer)
Context IR	Optional per-call surcharge when enhanced understanding is enabled (see model details)

Modes & ratio

Mode	When	`ratio`
t2va	text only	required, not `adaptive`
i2va	first/last frame	forced `adaptive`
r2va	`reference_*`	optional (default `adaptive`)

Exclusive: frames vs reference_*.
Equivalents: first_frame_url ≈ first_frame ≈ image ≈ images[0].

Create examples

Text-to-video, first-frame, first+last frame, multimodal native content, multimodal flat fields — see endpoint example dropdowns.