Videos Generated by Chinese AI Models Have Startled Us All

China's low-cost GenAI breakthroughs signal a will to leap ahead

- CATHERINE THORBECKE

Chinese AI companies are laser-focused on closing the development gap with the U.S. Despite Washington's efforts to hold the industry back, it's proving it can stay competitive with Silicon Valley. I spent some time playing around with Vidu 2.0, a revamped AI video generator that has been dubbed a domestic rival to OpenAI's Sora. Released by Shengshu Technology, a Beijing-based startup with ties to Tsinghua University, the public platform lets anyone from around the world turn images into short videos.

There were still notable inconsistencies in some of the clips I created: wonky facial expressions, limb movements that seemed to defy the laws of physics and other clear indications that these were AI-generated. These limitations also seem to plague Sora. But I was most impressed by its speed. In a matter of seconds, I was able to create fake clips of Donald Trump crying. The company says the real breakthrough is its ability to cut costs, claiming its short-form video content is produced 55% cheaper than the industry average.