MiniMax MCP
Generate MiniMax H3 multimodal videos in the AI Agent through MCP and manage asynchronous tasks.
MiniMax H3 · Multi-modal Video Generation API
MiniMax H3 Video Generation
Generate 4–15 second videos from text, up to 9 reference images, or up to 3 reference audio tracks using a REST API. Unified asynchronous tasks, callbacks, billing, and CDN delivery, directly integrated into your content products.
Text-to-Video / Image-to-Video / Audio-Guided768P / 2KAsynchronous Tasks + Webhook
3
Video Generation Modes
4-15s
Optional Video Duration
1-9
Reference Images
1-3
Reference Audio
One Interface, Three Creative Starting Points
No actions or multiple paths required; the interface automatically determines the generation mode based on content materials.
Text-to-Video
Directly generate short films from scene, action, shot, and style descriptions.
Frame-to-Frame Video Generation
Control the start and end frames through first_frame and last_frame.
Multi-modal References
Combine reference images, videos, and audio to guide subjects, actions, sounds, and rhythms.
Asynchronous Delivery
Immediately return task ID and receive the final CDN video via polling or callback_url.
curl -X POST 'https://api.acedata.cloud/minimax/videos' \
-H 'Authorization: Bearer YOUR_API_KEY' \
-H 'Content-Type: application/json' \
-d '{
"model": "MiniMax-H3",
"content": [{
"type": "text",
"text": "A red fox running through a snowy forest at dawn, cinematic tracking shot"
}],
"resolution": "2K",
"ratio": "16:9",
"duration": 5
}'Start Generating with a Few Lines of Code
Parameters directly express creative intent, while the platform handles long task execution, result storage, and precise billing.
1
Construct Content
All modes must provide text; images, videos, and audio are labeled for use through corresponding types and roles.
2
Submit Asynchronous Task
The interface is fixed asynchronous; it immediately returns task_id upon success.
3
Obtain Permanent Results
Get AceDataCloud CDN video from task.content.url via `/minimax/tasks` or callback.
Common Scenarios from Material to Finished Film
The same model covers creative exploration, activation of static materials, and music rhythm short films.
Marketing Short Film
Quickly generate horizontal ads or vertical social media materials from product selling points and shot scripts.
Character Consistency
Constrain character, clothing, props, and scene style with multiple reference images.
Rhythm Visuals
Make dance, camera movement, and scene changes follow the rhythm of the reference audio.
Three Steps to Go Live
Utilize AceDataCloud's unified authentication and task system without the need to build a separate media pipeline.
01
Get Token
Create an application in the console and obtain a unified Bearer Token.
02
Submit Generation
Select duration and aspect ratio, and add text, images, or audio materials.
03
Deliver Video
Poll for tasks or receive callbacks to integrate CDN video into the product workflow.
Unified Video Interface for Production
Converge task execution, error semantics, usage records, and file delivery into a stable contract.
| Capability | AceDataCloud MiniMax H3 | Some Others |
|---|---|---|
| Multimodal Input | Unified entry for text, images, videos, and audio | Multiple interfaces stitched together |
| Long Tasks | Polling and Webhook | Requires self-built task queue |
| Result Files | Automatic storage on platform CDN | Inconsistent link lifecycle |
| Failure Billing | No charges for failures | Requires self-verification of bills |
Select Workflow by Content
The model is fixed to MiniMax-H3, and the mode is automatically inferred from the official V2 multimodal content array.
TEXT
text
Only transmit text content, directly generating visuals from ideas, scripts, and shot descriptions.
FRAME
first_frame / last_frame
Use the first frame or the first and last frames to control the start and end visuals of the video.
REFERENCE
reference media
Combine reference images, videos, and audio to control subjects, actions, sounds, and rhythms.
TASK
task_id / callback
Create fixed asynchronous tasks to obtain results through task queries or callbacks.
Transparent Resolution Pricing
Charged by video seconds, only for successful tasks.
768P
$0.057143 / second
- 4 seconds as low as $0.228572
- Suitable for quick creative validation
- Three input modes
2K
$0.091429 / second
- 4 seconds as low as $0.365716
- Default output resolution
- Asynchronous task query
Frequently Asked Questions
Key notes on modes, materials, duration, and task execution.
Is action required?
No. When content is only text, it is text-to-video; first_frame / last_frame is image-to-video; reference_* is multimodal reference video. All modes require non-empty text.
What ratios and durations are supported?
Resolution is required, supporting 768P and 2K; ratios support adaptive, 21:9, 16:9, 4:3, 1:1, 3:4, and 9:16; duration is an integer from 4 to 15 seconds.
How to obtain results?
Create an interface that is fixed asynchronous and immediately returns task_id; use `/minimax/tasks` to query, and on success, get the video from task.content.url, or provide callback_url.
Will failed tasks incur charges?
No. Only tasks that are successfully completed and return video results are charged based on final duration.
More Video Services
Combine other video capabilities based on model style and workflow needs.
Integrate MiniMax H3 into Your Product
One Token, one generation entry, a task system, start building multimodal video experiences.
Introduction
APIs
Pricing
MCP
Skill