Chinese AI company MiniMax has open-sourced its new H3 video generation model but imposed licensing restrictions that block free access for users in the United States, European Union, United Kingdom, and South Korea1.
After the Shanghai-based company released the model's weights to developers on Monday, users discovered that the license curtailed access in those major overseas markets. The weights — the underlying parameters that encode the model's capabilities — are available to developers elsewhere, but the geographic carve-outs mark an unusual approach to open-source AI distribution.
The restrictions underscore the copyright challenges facing generative video AI. One characterization of the move frames it not as censorship but as self-protection driven by Hollywood copyright lawsuits2.
ANALYSIS The jurisdictions MiniMax excluded — the US, EU, UK, and South Korea — are among the most active in enforcing intellectual-property claims against generative AI companies. By withholding access in those markets, MiniMax limits its legal exposure while still distributing the model globally.
The licensing structure creates a two-tier open-source release: developers outside the restricted jurisdictions can freely use H3's weights, while those inside face barriers. This approach differs from the all-or-nothing licensing typical of most open-weight model releases.
For AI labs and developers in the US and EU, the practical effect is exclusion from one of the latest Chinese video-generation models at a time when competition in AI-generated video is intensifying across multiple companies and geographies.