A startup based in China, Sand AI, has introduced an open-source video-generating AI model that has garnered significant praise from prominent entrepreneurs, including Kai-Fu Lee, the founding director of Microsoft Research Asia. However, it appears that Sand AI is censoring certain images that could potentially raise concerns with Chinese regulators on the hosted version of the model, according to testing conducted by TechCrunch.
Earlier in the week, Sand AI announced the launch of Magi-1, a model that generates videos by predicting sequences of frames in an autoregressive manner. The company claims that Magi-1 can produce high-quality, controllable footage that accurately captures physics more effectively than rival open models.
👀 You won’t believe this is AI-generated
🧠 You won’t believe it’s open-source
🎬 You won’t believe it’s FREEMagi-1 video model has outperformed commercial video tools
For more details and examples, click here: 👇 pic.twitter.com/zlXRecWeqH
— Farhan (@mhdfaran) April 21, 2025
Due to its massive size of 24 billion parameters, Magi-1 is impractical to run on most consumer hardware, requiring between four and eight Nvidia H100 GPUs. As a result, Sand AI’s platform is the primary means for many users, including this reporter, to test drive Magi-1.
To initiate video generation, the platform requires a “prompt” image. However, not all prompts are allowed, as TechCrunch discovered. Sand AI blocks image uploads of Xi Jinping, Tiananmen Square, Tank Man, the Taiwanese flag, and insignias supporting Hong Kong liberation. The filtering occurs at the image level, and renaming image files does not bypass the blocking.
Sand AI is not the only Chinese startup that prevents uploads of politically sensitive images to its video generation tool. Hailuo AI, a Shanghai-based generative media platform, also blocks photos of Xi Jinping. However, Sand AI’s filtering appears to be more aggressive, as Hailuo allows images of Tiananmen Square.
As explained by Wired in a January article, models in China are required to adhere to stringent information controls. A 2023 law prohibits models from generating content that “damages the unity of the country and social harmony,” meaning it counters the government’s historical and political narratives. To comply, Chinese startups often censor their models through prompt-level filters or fine-tuning.
Interestingly, while Chinese models tend to block political speech, they often have fewer filters than their American counterparts for pornographic content. A recent report by 404 revealed that several video generators released by Chinese companies lack basic guardrails that prevent people from generating nonconsensual nudity.
