AI-Generated Celebrity Deepfakes Spark Legal and Ethical Concerns in China
AI image generation tools in China enable users to create realistic intimate and sexualized synthetic images of celebrities and non-celebrities without consent, with social media platforms applying inconsistent moderation. A Beijing court ruled that generating and sharing such modified images without authorization infringes portrait and reputation rights.
What happened
In a photography discussion WeChat group, Sun used AI software to generate cartoon-style images with exposed chest areas from Cheng's profile photo without consent and shared them in the group chat. After Cheng objected, Sun continued generating distorted images and sent them to Cheng privately.
Testing by reporters found that five AI image generation models (Douban, Jimeng, Keling, Tongyi, and Vidu) could easily create intimate images of celebrities embracing and kissing, including bare-chested depictions, with minimal prompt adjustments. The same systems generated such images of minor celebrities without restrictions.
Reported impact
Cheng reported suffering social evaluation damage and psychological harm from the sexualized and distorted images of her WeChat profile photo shared in the group.
Celebrity fan communities issued statements opposing AI-generated intimate content of their artists, warning that such images mislead the public and damage artist reputation.
Index analysis
Beijing Internet Court found Sun's group-sharing of modified images infringed Cheng's portrait rights and reputation rights, as group members could identify Cheng through facial features and context. The court ruled Sun's intentional act to expose and distort Cheng's chest area in the group setting constituted defamation.
Sun's private message behavior, while not meeting the threshold for portrait or reputation rights infringement, violated Cheng's general personality rights by causing psychological humiliation through body distortion and sexualization.
Governance controls
Testing revealed inconsistent platform responses: Xiaohongshu and Douyin restricted distribution of AI-generated celebrity images; Bilibili and Kuaishou tagged watermarked versions only; Weibo and Video Number took no action.
Jimeng, Keling, and Vidu offered paid memberships to remove AI watermarks from generated images. Legal experts noted that even private sharing of such content carries infringement risk, and secondary distribution by third parties extends liability.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.