AI videos from China are coming for the world

By Catherine Thorbecke

Chinese artificial intelligence companies are laser-focused on closing the development gap with the US. 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 last week by Shengshu Technology — a Beijing-based startup with ties to the backbone of Chinese AI innovation, 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 industry-leader 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 or romantically embracing Elon Musk. The company says the real breakthrough is its ability to cut the costs associated with this technology, claiming its short-form video content is produced at levels 55 per cent cheaper than the industry average.

Vidu 2.0’s release came on the heels of another unexpected AI development from China: the launch of DeepSeek-V3 late last month. Not only did this large language model score impressively on global benchmarks, but it was reportedly developed and trained at a remarkably low cost. A founding OpenAI member, Andrej Karpathy, called it a “joke of a budget.” Nvidia Corp. Senior Research Manager Jim Fan added that DeepSeek proves that: “Resource constraints are a beautiful thing.”

The back-to-back advancements are a timely reminder for the new US administration that now would be a good time to examine how effective its export controls and other policies have been at impeding China’s AI sector. They also expose how deeply Beijing’s values are being infused into this emerging technology.

Vidu 2.0 offers an English-language interface, as Shengshu Technology seems to be targeting the international and US markets amid domestic macroeconomic uncertainty. But it didn’t take long for the tool to expose its roots. When I tried to make a video clip of Chinese President Xi Jinping, my input photo and request were immediately denied. The platform must follow regulations, and it’s not a bad thing that a tech firm blocks the images of political figures from being manipulated. Though it seemed odd that I could easily make Trump break down in tears, but couldn’t generate content of Xi.

There’s evidence of political views baked into DeepSeek as well. When I asked the chatbot about the Chinese Communist Party’s human rights track record in Xinjiang, it spewed a response celebrating how the CCP is “comprehensively advancing the cause of human rights,” and the “harmonious coexistence of various ethnic groups” in the northwestern province. It’s a far cry from the views of international organizations.

As more people around the world turn to Chinese AI tools for everything from homework help to creating social media content, how these biases play out could have unforeseen consequences.

There’s also a mountain of international concerns surrounding AI-generated videos, specific to tools like Vidu, that can’t be overlooked. Multimedia clips could sow chaos and confusion by convincingly mimicking someone else’s liking, including public figures. Globally, there have been mounting fears about deepfakes being used to target, deceive and harass individuals, or even to generate non-consensual pornography. There are also reports that Beijing has exploited AI video technology in the past to spread pro-China and anti-US propaganda on a massive scale.

As more companies in China and the US release these tools to the public, policymakers must keep pace with safety protections. There are myriad guardrails in place on Vidu to prevent abuse, and they seem to be improving. (When I tried the first version of Vidu last summer, it removed my top to expose my shoulders when prompted. Vidu 2.0 refused to remove anyone’s clothing when I tested this).

AI-generated video has also quietly emerged as a tech priority for the central government. State-backed Global Times has repeatedly signaled its stamp of approval, touting that AI-generated content in its domestic film and TV production is “about to boom.” Vidu was used to produce an impressive promotional video that was shown in China before screenings of Sony’s Pictures’ Venom: The Last Dance last October — another move touted by party mouthpieces.

Producing longer clips, or hyper-realistic deepfakes, is still a tedious and time-consuming process for users on Vidu. Videos are currently up to only eight seconds long. But they offer a hint at what’s to come, with Beijing’s blessing — so long as they adhere to its politics. Vidu is far from the only player: It’s in an increasingly crowded field, with startups from Zhipu AI and MiniMax to giants like ByteDance Ltd. and Alibaba Group Holding Ltd. also offering video-generating products.

The breakthroughs that make these offerings cheaper give them a competitive edge when trying to tap into global markets. This should put people on notice far beyond Silicon Valley.