Do you think plagiarism is still just a superficial copy?
Now, even the underlying parameters and structures are under strict legal scrutiny.
For short-video and selfie app developers, a popular "anime face" filter can quickly make a platform go viral among young users.
What you may not know is that the hit "manga transformation effect" relies entirely on a heavily invested and extensively trained AI neural network model.
Recently, the People’s Court of Chaoyang District, Beijing, announced the first AI model infringement case in China, sending a clear message: plagiarism will no longer pay off.
I. An AI Lawsuit Sparked by a Special Effect
In June 2020, Douyin launched a "Manga Transformation Effect". With a single selfie, the tool could real-time convert portraits into 2D anime-style faces. It became an instant hit, amassing over 17 million users and 290 million plays in just six months.
However, two months later, a nearly identical "Girl Manga Effect" appeared on another selfie camera app.
Douyin found the outline, eye highlights, and hair texture of the two effects were almost identical.
Douyin claimed its heavily trained model structure and parameters were improperly copied, and filed a lawsuit against the defendant.
II. Court Rules on Parameters, Not Just Visual Effects
The defendant argued that "similar technical principles lead to similar effects", but the court rejected this defense.
The core issue of the case was: does the law protect the underlying data and parameters copied directly from the code?
It is well known that the Copyright Law protects expression rather than ideas. The video effect is an "expression", while the AI parameters guiding the transformation are more like a method or formula. Before this case, protecting such parameters under copyright law faced major legal obstacles.
The court found that Douyin’s parameters were highly valuable, developed through massive data training and continuous optimization, representing huge R&D efforts and generating substantial commercial benefits.
Technical appraisal confirmed that 33 out of 36 convolutional layers in the defendant’s model had identical parameters, with a core similarity of 91.7%. Such a high degree of resemblance could not result from independent research, proving direct copying.
The court departed from the traditional Copyright Law framework and ruled that the "Manga Transformation Effect" constitutes competitive interest protected by the Anti-Unfair Competition Law.
It explicitly stated:
"The parameters and structure of an AI model formed by an operator through data training and optimization constitute competitive interests protected by the Anti-Unfair Competition Law, as they bring innovative advantages and business benefits."
The defendant’s act amounted to misappropriation of Douyin’s innovative achievements, creating "substantial substitution" and disrupting fair market competition.
The final second-instance judgment ordered the defendant to pay 1.6 million yuan in compensation to Douyin.
III. Key Warnings for Entrepreneurs and Managers
This 1.6-million-yuan judgment serves as a critical survival guide for entrepreneurs.
For AI startups, patents and software copyrights remain foundational protections. This precedent also highlights the need to safeguard your model’s core parameters. The law will no longer tolerate improper "reverse engineering" and plagiarism.
For legal and industry professionals, this case offers practical guidance: if your data assets cannot be fully protected by copyright law, you may claim rights based on infringed competitive interests, a viable new path for protecting new-quality productive forces.
The court did not ban similar effects, but prohibited obtaining them through theft.
Protect Your Technological Moat
For apps with models embedded in clients, backend data is vulnerable to reverse extraction. Competitors can technically extract and decrypt models to obtain core parameters. R&D teams must build multi-layered security defenses combining technology and law. Protecting innovation requires safeguarding underlying data, not just optimizing interfaces.
Conclusion
While 1.6 million yuan may seem insignificant to tech giants, the message is clear:
In the AI era, copying parameters to chase trends is no longer feasible. In the age of new-quality productive forces, your code, data, and painstakingly tuned parameters are your core assets.
When others attempt to steal your achievements through improper means instead of independent research, you can trust that the courts and the industry will stand firmly behind innovation.
