A china -based boot, AI sandIt has released an AI model that is openly created with a license, which has gathered praise from businessmen such as Microsoft Research Asia’s establishment by Director Kai-Fu Lee. But Sand AI seems to censor images that could increase the rage of Chinese regulators from the hosting version of the model, according to TechCrunch.
Earlier this week, Sand AI announced MAG-1A model that produces “self -regulating” videos predicting sequences of frames. The company claims that the model can create high quality, controlled videos that captures physics more accurately than competing open models.
👀 You won’t believe this is ai created
🧠 You won’t believe it is an open source
🎬 You won’t believe it’s freeVideo Magi-1 Simply humiliated Video Commercial Tools
Details and examples below: 👇 pic.twitter.com/zlxReqh
– Farhan (@mhdfaran) April 21 2025
Magi-1 is too impossible to run in most consumer material. They are 24 billion parameters in size, and requires between four and eight NVIDIA H100 GPUS to run. (The parameters are the models of internal variables they use to make predictions.) For many users-this journalist includes-The Sand AI platform is the only place they can try Magi-1.
The platform needs a “quick” image to start video production. Not all prompts are allowed, TechCrunch quickly discovered. The Sand AI blocks the images of Xi Jinping, Tiananman Square and Tank Man, the Taiwan flag and the insignia that supports the release of Hong Kong. Filtering appears to be at the level of the image. Renaming the image files does not skirt.
The Sand AI is not the only Chinese starting that prevents the politically sensitive images to the video creation tool. Hailuo AI, the Hailuo AI media genetic platform, blocked XI Jinping photos as well. But Sand AI’s filtering seems to be particularly aggressive. Hauiluo allows images of Tiananmen Square.
As explained in a piece from January, models in China are required to follow strict information checks. A 2023 law prohibits models from creating content that “destroys the unity of the country and social harmony” – that is, it measures the historical and political narratives of the government. To comply, Chinese newly established companies often censor their models, either through direct -level filters or by refinement.
Interestingly, while Chinese models tend to prevent political speech, they often have fewer filters than their American counterparts for pornographic content. 404 recently reported That several video generators released by Chinese companies do not have basic protective messages that prevent people from creating non -consensual naked.
