Tom Cruise and Brad Pitt tussle in hand-to-hand combat on a rubble-strewn rooftop; Donald Trump takes on kung-fu fighters in a bamboo grove; Kanye West dances through a Chinese imperial palace while singing in Mandarin.

Over the past week, a slew of cinematic videos of celebrities and characters in absurd situations have gone viral online, with one commonality –– they were created using a new artificial intelligence tool from Chinese developer ByteDance, sparking anxiety over the fast-evolving capabilities of AI.

The new model, named Seedance 2.0, is among the most advanced of its kind and has quickly drawn praise for its ease of use and the realistic nature of the videos it can generate in minutes.

But soon after the release, media behemoths Paramount and Disney sent cease-and-desist letters to ByteDance –– the company most famous for developing the video-sharing app TikTok –– accusing it of infringing upon their intellectual property. Hollywood’s premier trade organization, the Motion Picture Association, and labor union SAG-AFTRA also condemned the company for unauthorized use of US-copyrighted works.

ByteDance responded with a statement saying it would implement better safeguards to protect intellectual property.

Seedance 2.0 has quickly become the most controversial model in a wave of them released by Chinese technology companies this year, as the competition to dominate the AI industry heats up.

China’s government has made advanced tech a key tenet of its national development strategy. In a televised Lunar New Year celebration this week, the country’s latest humanoid robots stole the show by performing martial arts, spin kicks and back flips.

Such improvements are often met with unease, particularly in the US, China’s chief technological and political rival, in a spiral of one-upmanship redolent of its 20th-century “Space Race” with the Soviet Union.

“There’s a kind of nationalist fervor around who’s going to ‘win’ the space race of AI,” said Ramesh Srinivasan, a professor of information studies at the University of California, Los Angeles. “That is part of what we are seeing play out again and again and again when it comes to this news as it breaks.”

Here’s why the latest technology from ByteDance has rattled the world.

The AI video generation model, while still not publicly available to everyone, was hailed by many as the most sophisticated of its kind to date, using images, audio, video and text prompts to quickly churn out short scenes with polished characters and motion editing control at lower cost.

“My glass half empty view is that Hollywood is about to be revolutionized/decimated,” writer and producer Rhett Reese, who worked on the Deadpool movie franchise, wrote on X after seeing the video of Cruise and Pitt.

One Chinese tech blogger using Seedance 2.0 said it was so advanced that it was able to generate realistic audio of his voice based solely on an image of him, raising fears over deepfakes and privacy. Afterwards, ByteDance rolled back that feature and introduced verification requirements for users who want to create digital avatars with their own images and audio, according to Chinese media.

Rogier Creemers, an assistant professor at Leiden University in the Netherlands, who researches China’s domestic tech policy, said part of the concern stems from the rapid rate at which Chinese companies have released new iterations of AI technology this year.

That has also put China on the back foot in assessing the potential negative impacts of each improvement, he said.

“The more capable these apps become, automatically, the more potentially harmful they become,” said Creemers. “It’s a little bit like a car. If you build a car that can drive faster, that gets you where you need to be a lot more quickly, but it also means that you can crash faster.”

After outcry from Hollywood, ByteDance said in a statement that it respects intellectual property rights and will strengthen safeguards against the unauthorized use of intellectual property and likenesses on its platform, though it did not specify how.

User complaints prompted the recent ByteDance rollback and have also forced popular Chinese Instagram-like app RedNote to restrict any AI-made content that has not been properly labeled.

And the arrival of Seedance 2.0 coincides with a tightening of regulations for AI content in China.

China’s domestic regulation of AI surpasses the efforts of most other countries in the world, in part because of its longstanding censorship apparatus. Last week, the Cyberspace Administration of China said it was cracking down on unlabeled AI-generated content, penalizing more than 13,000 accounts and removing hundreds of thousands of posts.

However, the restrictions on AI-generated content on the Chinese internet are often unevenly enforced, Nick Corvino wrote in ChinaTalk, a China-focused newsletter. He attributed the problem in part to difficulties policing content across different apps, as well as incentives for tech companies to encourage user content.

“With Chinese social media platforms locked in fierce competition, both with each other and the Western market, none wants to be the strictest enforcer while others let content flow freely,” he said in a post following the launch of Seedance 2.0.

According to analysts, China is walking a fine line between encouraging domestic development of AI models and maintaining strict controls on how those models are used.

“People in the AI business would always say what the Chinese government is doing is slowing down the development of AI,” said Creemers of Leiden University. “Obviously a content control system like the Chinese that essentially limits what you can produce, that’s never fun.”

Pressure to stop using certain images or data, from US media giants or other sources, may also impact efforts to refine AI. Disney accused ByteDance of illegally using its IP to train Seedance 2.0, but recently struck a deal with US company OpenAI to give Sora – OpenAI’s video generation model and Seedance competitor – access to trademarked characters like Mickey and Minnie Mouse.

“These agreements have everything to do with what kind of data are they going to get access to that they would not have otherwise, or that their competitors would not have?” said Srinivasan from UCLA. “There’s a high probability that the Sora products could be more refined and more advanced, if the data are better suited for the models to learn from.”

At the same time, restrictions on how AI can be used or trained could also spur greater innovation, he said, noting how Chinese company DeepSeek –– blessed with a much smaller budget than the industry leaders –– built a competitive AI-powered chatbot.

“When it comes to Chinese breakthroughs in AI, the DeepSeek revelation was so important because they showed that there are other ways of training language models in ways that are more economical,” he said.