MiniMax-H3

MiniMax-H3 is MiniMax’s multimodal video generation model. Submit an asynchronous task to POST /minimax/v2/video_generation, then poll GET /minimax/v2/query/video_generation/{task_id} until the task finishes.

The /minimax prefix is required on AnyFast. The routes without this prefix are MiniMax’s upstream routes, not the AnyFast routes shown on this page.

Key capabilities

Text to video — Generate a video from one text prompt.
Frame-guided image to video — Supply a first frame, a last frame, or both.
Multimodal reference to video — Combine reference images, videos, and audio with a prompt.
768P and 2K output — Select either output resolution.
Generated audio — Generated MP4 outputs can include an audio track.
Asynchronous processing — Create once, then poll the returned task ID.

Model ID

Use the case-sensitive model ID MiniMax-H3.

For specific content requests, select the MiniMax-H3-nsfw model under the Direct resource group. The NSFW model ID adds the -nsfw suffix to the standard model ID.

Quickstart

Create a 768P text-to-video task:

cURL

curl --request POST \
  --url https://www.anyfast.ai/minimax/v2/video_generation \
  --header 'Authorization: Bearer YOUR_API_KEY' \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "MiniMax-H3",
    "content": [
      {
        "type": "text",
        "text": "A red paper airplane glides above the shoreline at sunrise as the camera follows from a low angle."
      }
    ],
    "resolution": "768P",
    "duration": 4,
    "ratio": "21:9",
    "aigc_watermark": false
  }'

The response contains a task ID:

{
  "task_id": "asyntask_example123"
}

Poll the task ID. Polling does not create another generation task.

cURL

curl --request GET \
  --url https://www.anyfast.ai/minimax/v2/query/video_generation/asyntask_example123 \
  --header 'Authorization: Bearer YOUR_API_KEY'

Every accepted POST creates a separate generation task and may be billed separately. Do not retry a create request blindly. Save the returned task_id and retry only the GET query while the task is queued or running.

Choose an input workflow

Text to video
Image to video
Reference to video

Include exactly one non-empty text item. Set a concrete ratio; adaptive is not valid for text-only generation.

{
  "model": "MiniMax-H3",
  "content": [
    {
      "type": "text",
      "text": "A quiet observatory opens its dome beneath a meteor shower."
    }
  ],
  "resolution": "2K",
  "duration": 5,
  "ratio": "16:9"
}

Add one first_frame or last_frame image, or one of each. The output ratio follows the input images and is reported as adaptive; other supplied ratio values are ignored.

{
  "model": "MiniMax-H3",
  "content": [
    {
      "type": "text",
      "text": "The subject turns naturally toward the camera as the light changes softly."
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "https://example.com/first-frame.png"
      },
      "role": "first_frame"
    }
  ],
  "resolution": "2K",
  "duration": 4,
  "ratio": "adaptive"
}

Use reference_image, reference_video, and reference_audio roles to guide identity, composition, motion, camera movement, or voice. Include at least one reference image or video; reference audio cannot be the only media input.

{
  "model": "MiniMax-H3",
  "content": [
    {
      "type": "text",
      "text": "Keep the character and wardrobe from Image 1, and follow the camera rhythm of the reference video."
    },
    {
      "type": "image_url",
      "image_url": {
        "url": "https://example.com/character.png"
      },
      "role": "reference_image"
    },
    {
      "type": "video_url",
      "video_url": {
        "url": "https://example.com/camera-reference.mp4"
      },
      "role": "reference_video"
    }
  ],
  "resolution": "768P",
  "duration": 4,
  "ratio": "16:9"
}

Request parameters

Parameter	Required	Supported values and behavior
`model`	Yes	Case-sensitive value `MiniMax-H3`
`content`	Yes	Array with exactly one non-empty text prompt plus optional media parts
`resolution`	Yes	`768P` or `2K`
`duration`	Yes	Integer from `4` through `15` seconds
`ratio`	Conditional	`21:9`, `16:9`, `4:3`, `1:1`, `3:4`, or `9:16`; `adaptive` is also supported outside text-only generation
`callback_url`	No	Public callback URL that can complete MiniMax’s challenge verification within 3 seconds
`aigc_watermark`	No	Boolean; defaults to `false`

The text prompt can contain up to 7,000 characters. Send aigc_watermark as a JSON boolean; null, strings, and numbers are not part of the documented contract.

Media limits

Use publicly reachable HTTP or HTTPS URLs. Base64 media input is not supported. MiniMax must be able to download each URL while processing the task.

Media	Formats	Limits
Image	JPG, JPEG, PNG, WEBP, HEIC, HEIF	Up to 30 MB each; width and height 256-5,760 px; aspect ratio 0.4-2.5; up to one first frame, one last frame, or nine reference images
Reference video	MP4 or MOV; H.264/H.265 video; AAC/MP3 audio	Up to 50 MB each; 2-15 seconds each; at most 3 videos and 15 seconds total; width and height 256-5,760 px; aspect ratio 0.4-2.5; 23.976-60 fps
Reference audio	WAV or MP3	Up to 15 MB each; 2-15 seconds each; at most 3 audio files and 15 seconds total

Keep the combined number of image, video, and audio files at 12 or fewer. Do not mix first_frame or last_frame with any reference_* role.

A create request can return a task_id before MiniMax finishes downloading and validating remote media. An unreachable URL or a reference video even slightly longer than 15 seconds can therefore fail asynchronously. Continue polling until a terminal status instead of treating task creation as generation success.

Query task status

The task progresses through queued and running, then ends as succeeded, failed, or cancelled. When it succeeds, download the generated video from task.content.url. The URL may expire, so copy the file to your own storage promptly. The task.usage object describes processed media:

output_seconds is the generated duration used by the API.
input_seconds is the processed reference-video duration and may be omitted when no reference video is used.
input_image_count is the number of input images.
total_seconds is the sum of reported input and output seconds.

The requested duration and usage.output_seconds are integer seconds. A downloaded MP4 container can be slightly longer because of encoded frame and audio boundaries.

API reference

View all request fields, response fields, and error formats for MiniMax-H3.

Getting started

Model Introduction

Resources

Key capabilities

Model ID

Quickstart

Choose an input workflow

Request parameters

Media limits

Query task status

API reference

Official references

​Key capabilities

​Model ID

​Quickstart

​Choose an input workflow

​Request parameters

​Media limits

​Query task status

API reference

​Official references

Key capabilities

Model ID

Quickstart

Choose an input workflow

Request parameters

Media limits

Query task status

Official references