China’s MiniMax curbs overseas access to new AI video model over copyright disputes

Chinese AI firm MiniMax has released its new H3 video model but imposed licensing restrictions on users in key overseas markets, such as the US and European Union, highlighting the copyright challenges in generative video AI. After unveiling the model’s weights to developers on Monday, users discovered that the license limited unrestricted access in the US, EU, UK, and South Korea.

The weights, which encode the AI's intelligence, enable developers to use, copy, modify, and distribute the software without incurring costs to MiniMax.

To ensure compliance with the licensing terms, developers in major markets must apply for a formal license. MiniMax stated that once it has reviewed deployment scenarios and confirmed appropriate compliance controls and safeguards, it will authorize usage. However, access via application programming interface, available through the firm’s website or authorized third parties, remains globally accessible. The regional restrictions were attributed to regulatory uncertainties and ongoing copyright-related legal proceedings.

The US restriction is linked to "ongoing generative video copyright litigation with major Hollywood studios," according to Ryan Lee, MiniMax's head of developer relations, who announced the decision on X. Open-source video models have become a focal point in generative AI, with Chinese developers competing against US rivals by offering models that can be customized and deployed locally.

MiniMax's H3 model, known for its advanced video generation capabilities and broad hardware compatibility, can create videos up to 15 seconds long at 2K resolution, priced at 13 US cents per second. It outperformed Google’s Gemini Omni Flash in benchmark testing, ranking first in video-editing capabilities and second in text-to-video capabilities, though Google's model is limited to 10-second clips and costs just 1 US cent per second.

The licensing requirement comes amid mounting copyright disputes between generative AI companies and content owners over the use of protected works in training AI models and generating outputs. In June, Walt Disney and NBCUniversal sued Midjourney, a San Francisco-based research lab, alleging that its AI service generated unauthorized versions of copyrighted characters from Star Wars, Marvel, and other franchises.

Earlier this year, Chinese tech giant ByteDance faced backlash from Hollywood following the launch of its Seedance 2.0 model, which was later updated to Seedance 2.5, capable of producing 30-second videos in a single run. However, Seedance 2.5 is not open-sourced.