China has released draft regulations on AI copyright infringement, amid growing reports from actors, social media influencers, and ordinary citizens that their likenesses have been stolen for use in AI short dramas.

Released April 3 by the Cyberspace Administration of China, the draft “Administrative Measures for Digital Virtual Human Information Services” is open for public commentary until May 6.

According to industry insiders, AI-generated videos can be created from just a few photos of a subject, with the total cost of an AI-generated short drama typically a few thousand yuan and production lasting less than five days. In contrast, top human actors often charge between 15,000 and 30,000 yuan per day ($2,195 to $4,330), and traditional production cycles require seven to 10 days of shooting, followed by an editing period.

AI production companies often directly collect images of actors from the internet as references during their creation process. The draft states that companies would first need to obtain consent from the individuals whose images they seek to use and, in cases involving minors, from their parents or guardians. It notes that if the individual withdraws consent at any point, companies must delete any related personal information used. The draft also requires companies to respect the legal, portrait, and reputational rights of those whose likenesses they use, forbidding caricatures and disparagement.

A day before the release of the draft regulations, the Actors Committee of the China Federation of Radio and Television Associations — a national organization that protects the legal rights of actors — released a statement condemning the unauthorized use of AI for voice cloning, or using AI to create a voice that sounds like a specific individual, and face swapping, or using AI to stitch one person’s face onto another’s body, declaring that stealing a performer’s likeness violates their legal rights.

The statement emphasizes that AI-generated content using others’ likenesses remains illegal even if labeled “non-commercial,” “shared for the public good,” or “derivative work.” Such self-declared labels cannot constitute a legal basis for exemption.

The committee also prohibited scraping existing film and television data to train AI models without actors’ consent and asked short video platforms to build stronger verification tools, police their own content, and immediately take down pirated content.

Hongguo, a short drama app owned by ByteDance, removed hundreds of videos in the first quarter of 2026 after discovering they relied on stolen AI likenesses. One well-known influencer, going by the handle “Baicai,” posted on lifestyle platform Xiaohongshu, also known as RedNote, that his likeness had been used in the popular short drama “Peach Blossom Hairpin,” released earlier this year. He stated that the show had pulled his image from New Year’s photos he posted publicly online on Xiaohongshu. The 72-episode short drama garnered more than 41 million views on Hongguo before it was finally taken down on March 31.

Major stars are also taking action. On April 6, the studios representing popular domestic actors Zhang Jingyi and Deng Wei issued formal statements alleging that their images are currently being used in Hongguo short dramas without authorization and are suspected of infringing on their portrait and reputation rights. They added that they have tasked lawyers with collecting evidence and initiating legal action.

Leading dubbing companies, such as 729 Voice Studio and Bianjiang Studio, have also previously publicly condemned AI copyright infringement. 

However, tech insiders say it can be hard to determine what exactly constitutes AI copyright infringement. The head of an AI technology company, speaking to domestic media on the condition of anonymity, explained how determining whether a voice has been stolen by AI is often a “percentage question” rather than a clear-cut “yes or no.” Even if voice actors are willing to spend tens of thousands of yuan on voiceprint verification, they may not get a definitive answer, the source added.

Editor: Marianne Gunnarsson.

(Header image: VCG)