MiniMax Request Parameters and Media

H3 content roles, resolution, duration, media limits, Context-IR, regeneration, and Max platform options.

Native H3 creation preserves content order, media roles, and explicit false / 0 values. The selected upstream model validates supported fields. This page combines the official H3 creation reference with the formats integrated into aiiqq.com; the overview lists the five supported models.

H3 generation and Context-IR

Field	Usage
`model`	`MiniMax-H3` on all three native creation paths; preserve capitalization
`content`	Required array containing a nonempty `type: "text"` item and any media
`resolution`	Explicitly set `768P` or `2K` for H3 generation; Context-IR needs no output resolution
`duration`	Integer target duration of `4`–`15` seconds for H3 generation and Context-IR
`ratio`	Text-only input requires a specific ratio; see media cases below
`callback_url`	Optional upstream-validated callback; H3 callback verification requires echoing the supplied `challenge` within three seconds. Poll with the public ID rather than an upstream callback ID
`aigc_watermark`	Optional boolean, for example `false`; do not pass a string

Specific ratios are 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16.

content structure and roles

`type`	Content field	Roles / use
`text`	`text`	Include a nonempty prompt in every request
`image_url`	`image_url.url`	`first_frame`, `last_frame`, or `reference_image`
`video_url`	`video_url.url`	`reference_video`; regeneration also uses `base_video`
`audio_url`	`audio_url.url`	`reference_audio`

Frame input and reference-media input are mutually exclusive. When using any reference_image, reference_video, or reference_audio, do not include first_frame / last_frame in the same request.

Input	`ratio`
Text only	Required specific ratio; do not use `adaptive`
First / last frame	Determined by the image and handled as `adaptive`
Multimodal references	Explicit ratio or `adaptive`

First and last frames

{
  "model": "MiniMax-H3",
  "content": [
    {"type": "text", "text": "The camera moves closer as a flower bud gradually blooms in morning sunlight."},
    {"type": "image_url", "image_url": {"url": "https://example.com/first.png"}, "role": "first_frame"},
    {"type": "image_url", "image_url": {"url": "https://example.com/last.png"}, "role": "last_frame"}
  ],
  "resolution": "768P",
  "duration": 4,
  "ratio": "adaptive"
}

Send this JSON to POST /v2/video_generation. Replace the example media URLs with directly downloadable resources.

Multimodal references

{
  "model": "MiniMax-H3",
  "content": [
    {"type": "text", "text": "Create a product film using the reference teapot, camera movement from the video, and rhythm of the audio."},
    {"type": "image_url", "image_url": {"url": "https://example.com/teapot.png"}, "role": "reference_image"},
    {"type": "video_url", "video_url": {"url": "https://example.com/camera.mp4"}, "role": "reference_video"},
    {"type": "audio_url", "audio_url": {"url": "https://example.com/music.mp3"}, "role": "reference_audio"}
  ],
  "resolution": "768P",
  "duration": 4,
  "ratio": "16:9"
}

H3 media specifications

This is a summary of official native H3 constraints; upstream validation is authoritative.

Media	Constraints
Images	JPG/JPEG/PNG/WEBP/HEIC/HEIF; ≤30 MB each; dimensions 256–5760 px; aspect ratio 0.4–2.5; at most one first frame, one last frame, and nine reference images
Video	MP4/MOV; H.264/H.265 with AAC/MP3 audio; ≤50 MB each; at most three clips, 2–15 seconds each and ≤15 seconds combined; dimensions 256–5760 px, aspect ratio 0.4–2.5, 23.976–60 fps
Audio	WAV/MP3; ≤15 MB each; at most three clips, 2–15 seconds each and ≤15 seconds combined

Use directly downloadable public HTTP(S) media URLs instead of Base64 bodies. The official total body limit is 64 MB; gateway ingress limits may also apply. aiiqq.com does not provide MiniMax file uploads or mm_file:// references.

Video regeneration

Send model: "MiniMax-H3" and resolution: "2K" to POST /v2/video_regeneration. The source must meet H3 768P output requirements. Choose one source method:

source_task_id: the public ID of a successful native-protocol H3 generation created by the same token within seven days, including eligible H3 tasks created through the generic interface. The original channel and supplier key are reused. The provider may require allowlisting for this method.
content: exactly one role: "base_video" item, the final prompt actually used for the source, and the same reference media.

The base_video needs an audio track, 24 fps, dimensions divisible by 32, and 107–362 frames. Max / Turbo platform tasks cannot be used as native source_task_id values.

{
  "model": "MiniMax-H3",
  "resolution": "2K",
  "content": [
    {"type": "text", "text": "The final prompt actually used to generate the source video."},
    {"type": "video_url", "video_url": {"url": "https://example.com/h3-768p.mp4"}, "role": "base_video"}
  ]
}

Max and Max Turbo platform format

Send these models to POST /v1/video/generate with model set to minimax-h3-max or minimax-h3-max-turbo. Max uses 480P / 768P and integer durations of 5–15 seconds; do not use the H3 2K or four-second examples.

Max supports extra.prompt_expansion_mode values disabled, balanced, and quality, defaulting to balanced. Put extra at the top level of a native platform request, or under metadata.extra in the generic format. The verified minimal Turbo request uses text, 480P, and five seconds; confirm other options with the upstream model rather than assuming all Max media capabilities apply.

{
  "model": "minimax-h3-max",
  "content": [{"type": "text", "text": "A product film of a red ceramic teapot in morning sunlight."}],
  "resolution": "480P",
  "duration": 5,
  "ratio": "16:9",
  "extra": {"prompt_expansion_mode": "balanced"}
}

task.id in platform creation and query responses is the aiiqq.com public ID. Read successful video URLs from task.outputs and final metering such as input/output seconds from task.metadata.usage.