China's AI Drama Boom Puts a 300 Yuan Price on Faces, With Risks That Can Outlast the Payment

A quick payment for a face that can keep working

For 300 yuan, a 29-year-old woman surnamed Li supplied personal information, took photographs and a short video on her phone, and signed a contract allowing her likeness to be used in artificial intelligence productions. She found the recruitment advertisement in a group for casual work. No audition or acting experience was required.

In Hebei, a mother named Li Yi described completing a similar process in less than half an hour. In Wuhan, a collection studio recorded more than 20 people during its first week. Ordinary faces are becoming production materials for China's expanding market in AI dramas, advertisements and other digital content.

The attraction is simple: someone can earn money without attending a film set. The difficult part is deciding exactly what the buyer receives. Permission for one character in one drama is a different bargain from permission to reuse a face across advertisements, pass it to other companies or train an AI model.

China already has rules governing this trade. The Personal Information Protection Law treats biometric information, including facial information used to identify people, as sensitive personal information. Businesses must meet stricter requirements when collecting and using it, even when someone accepts payment.

The market also offers less certain income than recruitment advertisements suggest. Some participants receive a few hundred yuan, while others upload photographs and wait months without a buyer. For those who do secure a deal, a small payment can come with a much longer argument about control.

Why producers are shopping for real faces

China released about 128,000 microdramas in the first quarter of 2026, according to figures attributed to the China Netcasting Services Association. More than 95% used AI. Microdramas are short episodic stories commonly made for viewing vertically on a mobile phone.

By the first eight months of 2026, the reported total had reached 430,000, with more than 90% using AI. The first 10 months brought nearly 550,000 releases, with AI involved in more than 94%. These figures describe an enormous production market, while the trade in licensed ordinary faces remains small.

A typical drama may need five or six principal characters and more than a dozen supporting characters. Producers want faces that remain recognizable as characters move between scenes. Viewers have also complained that generated characters look too similar, giving buyers another reason to seek distinctive human features.

Real photographs can provide a consistent reference for a character. ActID, a platform based in Shenzhen, lets buyers search a catalogue by age, appearance and production genre. Camille Yan, its chief marketing officer, described demand beyond conventional leading actors.

Producers want attractive faces for lead roles, but they also need distinctive ones like elderly people, for example,

Production company Sanxing offers another approach: carefully written prompts, varied expressions and repeated adjustments can create recognizable fictional characters without a real face. AI drama director Wendi Ma has also said that most productions still use fully generated characters. Buyers therefore have a choice between licensing a person and refining a synthetic face.

What sellers actually earn

ActID said that, by July 2026, it had around 800 registered users, about 300 of whom had agreed to license their images. Two productions had licensed approximately 10 faces. Its quoted prices ranged from 99 to 500 yuan per episode, with a 10% commission.

Chengdu platform New Claw, also known as Xinzhao, reported nearly 70 transactions worth about 40,000 yuan during its first two months. Dividing those figures gives an approximate transaction value of 570 yuan, before allowing for fees or differences between deals. The company has described a minimum price of 500 yuan per image.

Other platforms advertise around 100 yuan for an ordinary person's appearance in one short video, sometimes with a 500 yuan minimum. Some take approximately 30% of the payment, while others take 30% to 40%. Prices and commissions depend on the platform, the permitted use and whether the licence covers an episode, a video or a project.

Listings span 50 yuan to 99,999 yuan in the broadest advertised range. Professional actors commonly list higher prices, sometimes reaching 10,000 yuan. Actual ordinary transactions described by participants generally sit in the hundreds of yuan.

Wang Jinlin, a model in Guilin, listed her face on ActID in May and said by October that no studio had requested it. She expected a typical offer of 500 to 1,000 yuan if selected. ActID founder Liu Xiao described the service as an opportunity to be seen and chosen, rather than a guarantee of earnings.

Demand is concentrated more heavily on celebrities and established short drama performers. Actress Qi Wei, for example, has an AI avatar named Han Qingxia and serves as producer of the animated series featuring it. An ordinary person enters the market without the audience recognition that makes a famous face commercially valuable.

A photo upload can carry several permissions

Applicants may submit close photographs from the front and side, alongside photographs of their whole body from the front, side and back. Some recruitment processes also request short videos and a recording of specified speech. Identity verification connects that material to a named individual.

Platforms can offer restrictions covering villains, death scenes, romance or other scenarios. In one workflow, a buyer chooses a licence for a single use or project, the person has 24 hours to accept, and authorization documents are generated before the materials are released.

Li Yunkai, an intellectual property lawyer at Beijing Tianyuan Law Firm, warned that the wording can permit multiple productions, commercial uses, AI training or further licensing to other businesses. He described the resulting risk to participants.

"This could leave individuals with little control over whether their likenesses appear in misleading advertisements, inappropriate content or scams," he said.

New Claw has stated that permission for one project does not automatically extend to sequels, extra episodes or derivative works, and that its existing licences exclude model training. Those limits give a buyer permission to make specified content while keeping other uses outside the agreement.

Other agreements give platforms more control. One examined contract required sellers to offer the lowest price across all platforms, with a penalty of 10,000 yuan for each breach. Against a 300 yuan payment, that stated penalty is more than 33 times the initial fee.

What China's privacy law requires

The National People's Congress Standing Committee adopted the Personal Information Protection Law on August 20, 2021. It took effect on November 1 that year. Articles 28 and 29 require a specific purpose, sufficient necessity, strict protective measures and separate consent for processing sensitive personal information.

Separate consent means the person must specifically agree to the sensitive information use. A broad acceptance of general terms cannot substitute for a clear decision about handing over facial information for the proposed purpose.

Article 17 requires businesses to identify themselves and explain their purposes, methods, information categories, retention periods and procedures for exercising rights. Article 30 adds an explanation of why sensitive information is necessary and how its use may affect the person.

Jinan University law professor Liu Ying explains those duties in an official privacy commentary published by the Cyberspace Administration of China. The explanation says businesses must assess risks before processing sensitive personal information and retain the assessment and processing records for at least three years.

A separate official legislative explanation describes disclosure and consent as central protections, alongside limits on excessive collection. For a face licensing company, this makes the proposed use of every requested photograph, video and voice recording a compliance issue.

Children require particular care. Listings on one platform included two people aged 16 and one aged 17, raising questions about consent and contractual capacity. Under the privacy law, information concerning children younger than 14 is itself sensitive personal information, and processing requires consent from a parent or other guardian.

Can someone withdraw after signing?

Article 15 gives individuals the right to withdraw consent and requires an easy way to do so. Article 47 requires deletion in circumstances including withdrawal, unlawful processing and the expiry of the necessary retention period.

The law also addresses technical difficulty. Where deletion is technically difficult, or a legally required retention period has not expired, the business must stop processing except for storage and necessary security protection. A company must therefore explain how it will handle withdrawal, rather than leaving the issue out of the agreement.

AI training makes the practical task harder. Training changes a model using examples; a person's characteristics may influence the resulting system rather than remain solely in a folder of photographs. Removing the uploaded files and addressing information already incorporated into a model can require different technical steps.

The Civil Code provides another route concerning permission to use a portrait. Article 992 makes personality rights inalienable. A transaction grants permission to use someone's likeness while the underlying right stays with that person.

Article 1022 allows either party to end a portrait licence without a specified term after reasonable notice. Where a term exists, the person depicted may terminate for a justified reason, although compensation for resulting losses can arise under the law. Article 1021 directs disputed interpretations of portrait licence terms in favour of the person depicted.

Contract wording still matters. Agreed payment and permitted uses can create enforceable obligations. Courts can examine unfair standard terms and excessive penalties, while a participant may face a compensation claim when ending an agreement. Clear provisions on purpose, duration, further licensing and withdrawal reduce the space for disputes.

When a face appears without permission

On March 30, a Hanfu makeup artist surnamed Bai, known online as Baicai, received a screenshot from a friend who recognized him in the AI drama Taohua Zan. He said the production had used his face, clothing, makeup and accessories from photographs he had posted the previous year.

The 72 episode drama depicted the character Liu Da as idle and lecherous. The face appeared in three episodes. The platform displayed a popularity score exceeding 40 million. A model known as Qihai also alleged that the production used her likeness for an abusive villain, including a mole matching her own.

Hongguo removed the drama and suspended the producer from uploading episodes for 15 days. Bai said the character's face changed after he posted screenshots, and he later pursued legal action. The episode shows how quickly the visible material at the centre of a complaint can disappear or change.

Beijing Internet Court has applied public recognizability when deciding an AI face swapping dispute. A generated image can infringe portrait rights when viewers can identify the person despite alterations. In one case, the producer and broadcasting platform were ordered to apologize and pay compensation; the platform's copyright permission did not cover the separate portrait rights.

Guangzhou Internet Court has heard approximately 700 cases involving AI facial misuse over three years. Shanghai model Xu Fang, who has pursued a claim over alleged unauthorized use of his face on fashion products, explained why he would refuse a licence himself.

If I license my face, I have no idea how AI might eventually change it or whether it could portray me in degrading ways,

A crackdown reaches into production

The National Radio and Television Administration has scheduled a new campaign from November 1 through December 2026 targeting vulgar and repetitive microdramas, including unauthorized use of faces and voices. It is the third campaign targeting the sector during the year.

The regulator said 75,000 dramas violating rules had been removed and more than 1,500 accounts sanctioned during 2026. ByteDance separately reported removing more than 85,000 videos involving unauthorized AI reproductions of faces and voices since the start of the year.

The regulatory sequence spans several dates:

• January 2026: regulators targeted harmful AI alterations of established television works, including historical and revolutionary subjects.
• June 2026: a campaign targeted harmful content, vulgar material and copyright infringement.
• July 2026: the Microdrama Development Management Measures were published.
• September 1, 2026: the measures took effect, including visible AI notices in each episode.
• November 1 to the end of December 2026: the latest campaign targets content, facial and voice misuse, repeat uploads and recommendation practices.

The measures also require broadcasters to review and assess recommendation algorithms and prohibit models that encourage addiction or excessive spending. The latest campaign targets repeated uploads under changed titles or character names, alongside repetitive recommendations on homepages and popularity lists.

A proposed regulation on digital virtual human information services would address sensitive information used for modelling and image generation, including deletion after consent is withdrawn. Its final form and implementation date have not been specified. Existing privacy and portrait rights already govern the collection and use of personal likenesses.

Evidence matters when control is lost

People considering a licence can ask for the named project, allowed roles, duration, payment basis, commission and any permission for training or further licensing. They can also request a written explanation of who receives the files and what happens to them when permission ends.

Participants who have already signed can preserve the complete agreement, consent selections, messages and payment records. The privacy law gives them rights to access and copy personal information, request explanations of processing rules and seek deletion where the legal conditions apply.

Someone discovering an unauthorized appearance can record the route from the platform homepage to the drama, episode and scene, keeping the account name and publication details visible. Original photographs and messages from people who recognized the face can help document identification. A platform complaint can request removal and preservation of relevant records.

Financial security is another concern. Beijing's market regulator warned on October 8 that criminals collect photographs, videos and voices from social platforms to generate avatars and clone speech. Such material can support impersonation and scams. Suspected fraud can be reported to police, while unlawful personal information processing can be reported to the responsible regulators.

Xiao Lin, 25, considered licensing her face after seeing recruitment posts but decided against it because of security concerns. Her decision reflects the calculation facing other potential participants: the payment is immediate and measurable, while future use may be difficult to track. A clearly limited licence and a workable exit procedure are central to the value of the bargain.

Key Points

• Some recruits receive 300 yuan for photographs, video and permission to use their likeness in AI content.
• Ordinary licences commonly cost hundreds of yuan, but many registered users receive no requests.
• China's privacy law requires separate consent and strict safeguards for sensitive facial information.
• Permission for a project, further licensing and AI model training can involve different rights and conditions.
• Withdrawal and deletion rights exist, while information incorporated into AI models can be difficult to remove.
• The November to December 2026 regulatory campaign includes unauthorized use of faces and voices.
• Contracts, original photographs and recordings of disputed content can help people pursue complaints and legal claims.