Skip to main content
POST
Note for users in mainland China: Please use https://toapis.cn as the API endpoint (Base URL). Replace https://toapis.com with https://toapis.cn in the examples in this document.
  • Supports three models: MiniMax-H3, MiniMax-H3-Max, and MiniMax-H3-Max-Turbo
  • MiniMax-H3 and MiniMax-H3-Max support text-to-video, first/last-frame generation, and combined image, video, and audio references
  • MiniMax-H3-Max-Turbo supports only text-to-video and first/last-frame generation, without reference media
  • MiniMax-H3 keeps 2K / 768p and 4-15 seconds; Max / Max-Turbo support 480P / 768P and 5-15 seconds
  • Async API that returns a unified generation.task
MiniMax-H3-Max-Turbo does not support reference_image, reference_video, or reference_audio. Sending reference media returns MiniMax-H3-Max-Turbo does not support reference media. Use MiniMax-H3 or MiniMax-H3-Max for reference videos.
Prefer publicly accessible media URLs. First/last-frame mode and multimodal reference mode are mutually exclusive. Reference audio cannot be used alone and requires at least one reference image or video.

Authentication

string
required
All endpoints require Bearer Token authentication.

Request Parameters

string
default:"MiniMax-H3"
required
Video generation model name: MiniMax-H3, MiniMax-H3-Max, or MiniMax-H3-Max-Turbo. Multimodal references are supported only by MiniMax-H3 and MiniMax-H3-Max; Turbo supports only text-to-video and first/last-frame generation.
string
required
Video description, up to 7000 characters. Describe the subject, action, scene, camera movement, style, and sound requirements.
integer
default:5
MiniMax-H3 accepts 4 through 15; Max / Max-Turbo accept integer values from 5 through 15.
string
default:"2K"
MiniMax-H3 accepts 2K and 768p; Max / Max-Turbo accept 480P and 768P.
string
default:"16:9"
Output aspect ratio.Options: 21:9, 16:9, 4:3, 1:1, 3:4, 9:16, or adaptive.
  • Text-to-video: defaults to 16:9; adaptive is not allowed
  • First/last-frame mode: always adapts to the input image and ignores a specific ratio
  • Multimodal reference mode (MiniMax-H3 and MiniMax-H3-Max only): defaults to adaptive, or accepts a specific ratio
string[]
Image URL array for compatibility. New integrations should use image_with_roles to declare each image’s purpose explicitly.
  • 1 image: treated as first_frame
  • 2 images: treated as first_frame and last_frame in order
  • 3-9 images: treated as reference_image, supported only by MiniMax-H3 and MiniMax-H3-Max; not supported by Turbo
Do not send image_urls and image_with_roles together because image roles may become ambiguous.
array
Images with explicit roles.
array
Reference videos for multimodal reference mode, up to 3 videos. Supported only by MiniMax-H3 and MiniMax-H3-Max, not by Turbo.
ToAPIs reads the actual reference-video duration for billing. Do not submit a client-calculated duration field.
array
Reference audio for multimodal reference mode, up to 3 audio files. Supported only by MiniMax-H3 and MiniMax-H3-Max, not by Turbo.
Reference audio cannot be used alone. Include at least one reference_image or reference_video.
boolean
default:false
Whether to add an AIGC watermark to the generated video.
string
Your business identifier, such as an order or transaction ID. You can query the task using this value through the same status endpoint.
string
ToAPIs task-completion callback URL. Configure the Token URL and signing secret first; see Task Webhooks.

Input Modes

Multimodal reference limits (MiniMax-H3 and MiniMax-H3-Max only):
  • Up to 9 reference images
  • Up to 3 reference videos
  • Up to 3 reference audio files
  • Up to 12 media files total
  • First/last-frame roles cannot be mixed with reference roles

Request Examples

The reference-media examples use MiniMax-H3. You cannot use them by simply changing model to MiniMax-H3-Max-Turbo. To use MiniMax-H3-Max, set resolution to 480P or 768P and duration to 5-15 seconds.

Text-to-video

First and last frames

Multiple reference images

Combined video and audio references

Response

string
ToAPIs task ID used to query task status.
string
Object type, always generation.task.
string
Model used for this request, always MiniMax-H3.
string
Task status: queued, in_progress, completed, or failed.
integer
Task progress percentage from 0 to 100.
integer
Task creation time as a Unix timestamp in seconds.
After submission, poll Get Video Task Status. When completed, the video URL is available at result.data[0].url.