Celebrities face up to reality of AI image use

A subtle yet significant change is occurring in the world of AI-generated content involving celebrity likenesses.

Previously, public figures and celebrities opposed unauthorized use of their images, but now more of them are choosing to license their facial data for AI-generated content, signaling a notable shift in strategy.

In July, Joey Wong, the 59-year-old Taiwan actress who retired in 2004, made headlines by allowing Chinese gaming giant NetEase to use youthful images of her for Qianying, an 87-second AI-generated promotional trailer for the online game series Tianxia. The short film revives some of her iconic roles, such as the ethereal Nie Xiaoqian from A Chinese Ghost Story, with Wong not needing to appear on camera.

A few weeks earlier, 62-year-old Hong Kong actor Lawrence Ng announced that he had licensed his 20-year-old likeness for an AI-produced film. Ng described the fee as "considerable" and said he did not need to participate in the production. His contract strictly limited the use of his digital image to that specific project, according to media reports.

This new enthusiasm for cooperation contrasts sharply with events just a few months earlier. In late March, some bloggers, including models, traditional costume enthusiasts and actors including Yi Yangqianxi and Yang Zi, publicly expressed concerns over AI-generated content using their images and voices without permission. Their complaints drew significant public attention to the technology's use.

The shift from confrontation to cooperation has sparked debate about whether AI can extend celebrities' screen careers or merely replicate in digital form the appearances that made them famous.

Zheng Ning, head of the law department at the Communication University of China's Cultural Industries Management School, said that allowing actors to license their likeness for AI-assisted productions is an opportunity for veteran performers or those with physical difficulties to extend their careers.

The commercial appeal of such productions will help refresh classic works, through a reliance on nostalgia, retro imagery, and the audience's emotional attachment to veteran actors' earlier performances.

The shift toward proactive licensing is driven by more than just changing attitudes, she said. "It reflects technological maturity, booming industry demand, and a growing regulatory push for clearer consent rules," Zheng said.

Almost human, after all

Public figures' newfound enthusiasm for AI content producers using their images is not accidental.

For years, celebrities were on the defensive, chasing down unauthorized deepfakes — a costly, time-consuming and often futile effort. The rise of generative AI worsened the problem, as synthetic content became easier to produce and harder to regulate. This created a tense environment where celebrities had to be constantly vigilant, detracting from their creative pursuits.

Zheng said the shift happened as AI technology matured to a point where "high-fidelity digital humans" became commercially viable. Before then, many people felt uncomfortable encountering a digital representation of an image that was almost, but not quite, human. Overcoming this hurdle has been a significant milestone in AI development, enabling more realistic and acceptable digital doubles.

"The explosion of AI-generated short films and virtual content is creating genuine market demand for legally sourced facial data, while regulatory pressure is also mounting — rules now require explicit, separate consent for using sensitive biometric data in virtual human creation," Zheng explained.

A clear regulatory environment is crucial for responsible use of digital likenesses, ensuring creators and rights holders are adequately protected. She said this creates a clear incentive for both sides.

Content producers gain access to high-value, legally cleared material, while celebrities — particularly those who have retired or reduced their workload — can extend their careers without the physical demands of filming, by authorizing AI operators to use their likeness in new productions.

"They can also capture the economic value of the deep emotional attachment audiences still have to their classic roles," she said, adding that this emotional connection is a powerful driver in the entertainment industry, which AI technology can leverage in innovative ways.

However, not all members of the public are enthusiastic about such use. Some fans have said it makes them feel uncomfortable and that AI-generated avatars lack the "spirit" of the real performer.

"This ambivalence — excitement mixed with a lingering attachment to the authenticity of human performance — suggests that while the commercial logic is compelling, the emotional acceptance of digital doubles is still evolving," Zheng said.

This evolution reflects broader societal questions about authenticity and the role of technology in human experiences.

As the technology becomes more widely available, she said, the practice of licensing one's likeness to AI operators for use across a growing range of sectors is expected to reach everyday users. This "democratization" of digital identities could unlock new avenues for personal expression and creativity, opening up possibilities not only for individuals but also for entire industries. However, it also brings risks that cannot be overlooked.

Contract details crucial

Sun Mingxi, vice-president of the Beijing Internet Court, said that obtaining permission is the first step before personal information, especially sensitive data such as facial material, can be used in AI-powered tools.

"This legal foundation is essential for ensuring that digital likenesses are used ethically and responsibly, preventing unauthorized exploitation," Sun said.

In a case heard by the court in June 2024, a face-swapping app provider was fined for misusing the personal data of a popular model featured in videos with ancient Chinese themes. The plaintiff discovered her likeness was being used for free without her permission as a face-swapping template. The case underscored the importance of legal recourse in protecting individual rights in the digital realm.

In March, several bloggers, including traditional Chinese attire enthusiasts and models, accused a popular AI-generated series Peach Blossom Hairpin of using technology to replicate their facial features, outfits, and makeup without consent, to create characters that were widely distributed on short-video platforms. The drama was later pulled from online viewing.

While Sun stressed obtaining authorization or consent from users was a must for AI operators, she also said "it is not a 'get-out-of-jail-free' card".

She cited laws explaining that authorization for likeness use does not automatically include permission for AI model training.

"If a company uses authorized images to train a large model, that is a separate act of processing sensitive personal information and demands its own consent," she said. This requirement for separate consent reflects the complexity of AI technologies and the need for nuanced legal frameworks to address their specific challenges.

Tian Fen, a judge at the Beijing Chaoyang District People's Court, offered a checklist for crafting robust licensing agreements.

"Parties should clearly define whether the license is exclusive, specify which age range or source material is authorized, and restrict the types of AI products — for example, a specific film or game," she said.

"They should also explicitly prohibit use in derivatives, advertising, or data training, set clear time and geographic limits, and mandate data deletion protocols upon termination," she said. These guidelines provide a foundation for creating agreements that are clear, comprehensive, and enforceable, reducing the risk of disputes and misunderstandings, she added.

Zheng raised another critical concern: reputation risk.

"If AI generates dialogue, plots, or scenarios that were never performed — especially if they are vulgar, misleading, or damaging — this could trigger liability for defamation, unfair competition, or general personality rights violations," she warned.

"A celebrity might license their likeness, but they must retain script approval rights to ensure content quality and consistency with their public image."

The legal experts agree that generic, broad authorizations are a recipe for disputes. Instead, contracts should treat each use case separately, with clear boundaries that leave no room for interpretation, and with penalties for overreach that are severe enough to deter misuse.

This approach also emphasizes the importance of specificity and precision in legal agreements, ensuring that all parties understand their rights and obligations.

Lingering concerns

Despite the growing sophistication of licensing practices, significant legal voids remain.

"Once a model is trained, biometric data becomes woven into the model's parameters. If a contract expires, how do you 'delete' that training? There is no industry standard for technical verification of complete removal," Sun said.

This creates a lingering risk: even after a license ends, a digital likeness could theoretically persist within a model's architecture, waiting to be regenerated, she said.

Addressing this problem requires the development of new technical and legal standards that can effectively manage the life cycle of AI models and the data they contain, Sun said.

Tian, the judge from Chaoyang, also underscored the importance of clearly specifying in contracts the rules for deleting and destroying both the original facial material and the generated digital likenesses.

"The liabilities for breach, the calculation of compensation in the event of unilateral termination — all these must be spelled out up front, so that victims have a clear legal basis to protect their rights if things go wrong," she said.

When it comes to how the public should manage their image assets in the AI era, she noted that the pace of legal development inevitably lags behind the rapid advance of technology. "For emerging areas — such as the use of AI-generated derivatives, the licensing of digital likenesses after death, or the question of inheritance — clear rules have yet to take shape," she said.

Tian suggested the public cultivate a strong awareness of digital assets and insist on authorization before any use of personal data. She also called on them to strengthen their capacity for rights protection — including the ability to secure online evidence and preserve it immediately when infringement occurs, so they can seek legal redress without delay.

caoyin@chinadaily.com.cn