Chinese artificial intelligence startup MiniMax has launched H3, a new multimodal video-generation model capable of understanding and generating content from text, images, video, and audio, intensifying competition in China’s rapidly evolving AI industry. The Shanghai-based company said H3 can create 15-second videos in 2K resolution with native stereo sound, while also supporting advanced editing capabilities such as modifying existing videos and transferring motion between clips using multimodal prompts.
A notable feature of the launch is MiniMax’s commitment to release H3’s model weights within days, allowing developers to download, customize, and deploy the system. The move extends the growing open-weight trend in China’s AI ecosystem into video generation, a segment where many leading models remain proprietary. MiniMax also said H3 is optimized to run on several Chinese-made AI chips, supporting the country’s push to reduce dependence on U.S. semiconductor technology.
MiniMax Launches H3 Multimodal Video Model
H3 is designed as a general-purpose multimodal AI model capable of processing multiple input formats simultaneously.
Its capabilities include:
- Text-to-video generation.
- Image-to-video generation.
- Audio-guided video generation.
- Video editing using multimodal prompts.
- Motion transfer between videos.
- Native stereo sound generation.
According to MiniMax, the model is aimed at both creative professionals and enterprise users seeking AI-powered content generation.
H3 Feature Snapshot
| Feature | Details |
|---|---|
| Model | MiniMax H3 |
| Company | MiniMax |
| Input Types | Text, images, video, audio |
| Video Output | Up to 15 seconds |
| Resolution | 2K |
| Audio | Native stereo sound |
| Availability | Open-weight release planned |
Advanced Editing and Motion Transfer
Unlike many earlier AI video models that primarily generate videos from text prompts, H3 also enables users to edit existing videos.
Key editing capabilities include:
- Modifying existing video content.
- Transferring motion from one video to another.
- Combining multiple media formats within a single prompt.
- Generating synchronized video and stereo audio.
These multimodal workflows are designed to simplify professional content creation for industries such as advertising, gaming, and digital media.
Open-Weight Strategy Expands to Video AI
MiniMax has faced scrutiny over its AI video tools too, after it curbed overseas access to a new AI video model amid copyright disputes.
MiniMax said it plans to release H3’s model weights shortly after launch.
This approach allows developers to:
- Download and fine-tune the model.
- Build customized applications.
- Deploy H3 on their own infrastructure.
- Optimize the model for industry-specific use cases.
Open-weight AI models have become increasingly popular in China as companies seek to accelerate developer adoption and broaden enterprise deployment, contrasting with many closed-source video generation systems from global competitors.
Why the Open-Weight Release Matters
| Benefit | Impact |
|---|---|
| Customization | Developers can fine-tune the model |
| Enterprise Adoption | Easier deployment for commercial applications |
| Research | Encourages broader AI experimentation |
| Ecosystem Growth | Expands developer participation |
Targeting Commercial AI Applications
Competition in AI video generation is intensifying more broadly, with video-generation startup PixVerse raising $439 million at a valuation surpassing $2 billion.
MiniMax said H3 has been designed primarily for enterprise and commercial use cases, including:
- Advertising and marketing.
- E-commerce product videos.
- Product design visualization.
- Gaming content.
- Creative media production.
The company also claims that generating 2K videos costs less than one-third of comparable mainstream products, positioning H3 as a lower-cost alternative for businesses adopting generative AI.
Built for Chinese AI Hardware
As China accelerates efforts to develop domestic AI infrastructure, MiniMax emphasized that H3 is compatible with multiple Chinese-made AI chips.
The hardware optimization supports broader national efforts to:
- Reduce dependence on U.S. semiconductors.
- Improve domestic AI computing capabilities.
- Expand deployment across locally manufactured AI accelerators.
The announcement comes as Chinese AI companies increasingly optimize software for indigenous hardware platforms amid ongoing U.S. export restrictions on advanced chips.
Competition in China’s Video AI Market Intensifies
MiniMax joins a growing list of Chinese companies rapidly advancing AI video generation.
Recent competitors include:
- ByteDance’s Seedance 2.0.
- Kuaishou’s Kling 3.0.
The rapid pace of product launches highlights how Chinese AI developers are expanding beyond large language models into multimodal systems capable of generating text, images, audio, and video from unified prompts.
Looking Ahead
The launch of H3 reinforces MiniMax’s ambition to become one of China’s leading multimodal AI companies by extending its portfolio beyond language models into advanced video generation. With support for text, images, audio, and video, native stereo sound, sophisticated editing capabilities, and a planned open-weight release, H3 reflects the industry’s shift toward increasingly flexible and developer-friendly multimodal AI platforms.
Looking ahead, competition in AI video generation is expected to intensify as Chinese companies race to improve quality while lowering costs. MiniMax’s focus on commercial applications, open-weight distribution, and compatibility with domestically produced AI chips could strengthen its position among enterprise customers and developers, while contributing to China’s broader strategy of building a self-reliant AI ecosystem.
Frequently Asked Questions
What is MiniMax’s H3 model?
H3 is a new multimodal video-generation model from MiniMax that can understand and generate content from text, images, video, and audio.
What video quality can H3 produce?
H3 can create 15-second videos in 2K resolution with native stereo sound, and supports advanced editing capabilities such as modifying existing videos and transferring motion between clips.
Is H3 an open-weight model?
Yes, the release reflects MiniMax’s open-weight strategy expanding into video AI, according to the article.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
