Text Creation
Provide non-empty text in content, clearly specifying the subject, action, scene, and shot objective, and use one short video to express one clear task.
MiniMax H3 Max · Video
Create 480P and 768P short videos with text and reference materials
MiniMax-H3-Max uses unified content input and supports integer durations of 5–15 seconds. First select input materials and an output tier based on the task, then retrieve the completed result through platform task queries.
Specifications and API Features
- Model ID
- MiniMax-H3-Max
- API Endpoint
- POST /minimax/videos
- Input
- Text, images, video, or audio in content; see the API documentation for combination requirements
- Output Tiers
- 480P, 768P
- Duration
- Integer values from 5–15 seconds
- Delivery
- Asynchronous task_id, task queries, and optional callbacks
Core Capabilities
Reference Material Creation
Use first frames, last frames, or multimodal reference materials as documented. The type and role of different materials determine their purpose; do not use legacy top-level fields such as prompt and image_urls.
Flexible Short Video Duration
Schedule tasks in whole seconds within the 5–15 second range, choose a duration that matches the number of subject actions, and review the output after completion.
Use Cases
Creative Shot Previsualization
Use text or keyframes to discuss camera movement, composition, and pacing, providing dynamic reference for subsequent production.
Product and Event Assets
Use existing images as supported reference inputs, describe presentation actions and scene atmosphere, and generate short assets for later editing.
Reference-Driven Creation
When combining image, video, or audio materials, first clarify the reference purpose of each asset, then submit according to content combination limits.
How to Choose This Model
Choose by Output Tier
H3 Max provides 480P and 768P. For 2K, choose a model that supports that tier rather than submitting 2K to H3 Max.
Choose by Materials and Cost
Reference images and reference videos affect the cost structure. Refer to the Pricing page for current billing rules and package pricing.
Get Started
Prepare Content
Explicitly specify model=MiniMax-H3-Max, and provide text and any required reference materials in content.
Select Duration and Aspect Ratio
Choose 480P or 768P and an integer duration of 5–15 seconds. ratio requirements depend on the input workflow; configure it according to the API documentation.
Query the Same Task
Use async=true to save task_id, then query via /minimax/tasks, or use callback_url to receive completion notifications.
Usage Limits
- H3 Max does not support 2K, and the duration range is integer values from 5–15 seconds.
- The API uses the content structure and does not accept the legacy top-level prompt, image_urls, audio_urls, or messages fields.
- Audio input incurs no additional charge; the first two reference images are free, while additional images and reference video usage are billed according to current rules.
Frequently Asked Questions
How do I select H3 Max?
Submit a request to POST /minimax/videos, explicitly set model=MiniMax-H3-Max, and provide non-empty text in content.
What resolutions and durations are supported?
480P, 768P, and integer durations of 5–15 seconds are supported; 2K is not supported for this model.
How do I provide reference materials?
Use the content items and role values defined in the documentation to indicate first-frame, last-frame, or reference use, and comply with quantity and combination limits for each material type.
How are reference materials billed?
Audio input incurs no additional charge, and the first two images are free; additional images and reference videos are billed according to their respective input usage, while output video costs are calculated under current rules.
How do I retrieve the completed video?
An asynchronous request first returns task_id. Query the same task, and after confirming task.status=succeeded, retrieve task.content.url; if it fails, read the task error information.
Updated: 2026-10-01. Specifications and fees are subject to the current API and pricing pages.