A Chinese AI video startup appears to be blocking politically sensitive images

A basic start of China, You have the sandMicrosoft Research Asia, Kai-Fu Lee’s founding director, relied on entrepreneurs, released a video created a video. However, SUS AI, according to TechCrunch, the pictures that can lift images made of the hydoded version of Chinese regulators are censored by the homeowner version of the model to block blocked images.

The Sand AI was declared earlier this week Magi-1A model that creates videos by “autoregressely” that predicts the sequence of frames. The company claims that the model can create high quality and managed images that are more accurate to physics more accurately than opponent’s open models.

👀 You won’t believe this EU is created
🧠 You won’t believe the open source will not believe
🎬 You won’t believe it is free

Magi-1 video model only humiliated commercial video vehicles

The following details and examples: 👇 pic.twitter.com/zlxrecwegh

– Farhan (@mhdfaran) April 21, 2025

MAGI-1 is very impossible to work on most consumers. This is 24 billion parameter and required to escape from four to eight Nvidia H100 GPU. (Settings are internal variable models for predictions.) For many users – This correspondent includes – Sand Ai’s platform is the only place where they can test Magi-1.

The platform needs a “Quickly” image to prevent video generation, but all the instructions were not allowed, not allowed. Sand AI block loads support Sand AI, Xi Jinping, Tiananmen Square and Tank Man, Taiwan Flags and Hong Kong Salvation. The filter seems to be occurring at the image level – the change of image files was not blocked.

Sand AI is not the only start of China, which prevents the video offspring tool in the video generation of local generation. Hailuo EU, a Romanian-generative media platform, located in Shanghai, also blocks the photos of Xi Jinping. However, the sand EU filter looks especially aggressive; Hailuo allows the images of Tiananmen Square.

Like a string explained In one part in January, China’s models are required to follow serious information control. The 2023 Law prohibits the models that generate content generating content, which causes the country’s historical and political narratives, “to adapt to the adaptation of Chinese beginnings or correcting them or correcting the models or quickly.

Interestingly, Chinese models tend to block political speech, often have fewer filters for pornographic content than American colleagues. 404 Recently reported A number of video generators released by Chinese companies are the lack of major greenhouses that prevent other people’s consensual naked images.