Context: The issues surrounding AI and copyright have already stirred much debate in China, with one ruling handed down by the Beijing Internet Court – which found that AI-created artwork can be protected by copyright laws based on the “originality” and intellectual input of its human creator – setting a key precedent for future AI and copyright case law (November 28, 2023 Beijing Internet Court decision). In 2024, several more cases followed, including a major suit launched by Shanghai Xinchuanghua Cultural Development Co Ltd (Xinchuanghua) against a GenAI platform over the copyright infringement of a Japanese franchise character called “Ultraman”. The Guangzhou Internet Court sided with the plaintiff last March, specifying three key duties of AI service providers: notifying users via service agreements that they must not infringe upon others’ copyrights; establishing a complaint mechanism for rights holders to protect their copyrights; and providing prominent identification in case the AI-generated content could cause public confusion or misidentification.
What’s new: In a parallel action launched by Xinchuanghua before the Hangzhou Internet Court, the court yesterday ruled that GenAI-generated images of Ultraman on another GenAI platform constituted contributory copyright infringement (February 10, 2025 Hangzhou Internet Court decision). The court ordered the defendant to immediately stop the infringement and compensate Xinchuanghua with 30,000 Chinese yuan ($4,206). Interestingly, the court did not rule fully in Xinchuanghua’s favour, finding that the defendant does not provide training data for the AI itself but relies on users to upload images instead. This almost falls within the “safe harbor” rule for platforms, it found.
Direct impact: Following Xinchuanghua’s win in Guangzhou, the plaintiff could appeal against the decision to get the GenAI platform to pay for the full copyright infringement. However, given that the compensation is still so low in cases like this in China, it may not be worth it.
Wider ramifications: The suit may however serve as a reminder to GenAI platforms that even when users are the ones giving image prompts, they can be liable for the eventual copyright infringement that may follow via the AI outputs. As noted by the judge: “The defendant should have known that network users used its services to infringe upon the right of information network dissemination but did not take necessary prevention measures.” We are likely to see more such cases coming out of China soon.
Xinchuanghua owns the exclusive rights to reproduce, disseminate via information networks and adapt the Ultraman series (originally created by Japanese studio Tsuburaya Productions Co., Ltd.), as well as independently enforce those rights.
The unnamed defendant operates an AI platform that provides Low-Rank Adaptation (LoRA) models and supports many functions such as image generation and model online training. On the homepage of the platform and under “Recommendations” and “IP Works”, it offers users the opportunity to apply, download, publish or share AI-generated pictures and LoRA models related to Ultraman. This model was generated by users who uploaded Ultraman pictures, selected the platform’s basic model, and adjusted parameters for training. Afterwards, the platform allowed users to input prompts, select the base model, and overlay the Ultraman LoRA model to generate images that closely resembled the Ultraman character.
In its suit, Xinchuanghua alleged that the defendant infringed on its right to disseminate Ultriman and that it used AI technology to train the Ultraman LoRA model and generate infringing images.
However, the defendant argued that the platform does not provide training data and only allows users to upload images to train the model, which falls within the “safe harbor” rule for platforms. This does not constitute infringement, it said.
The court considered four main factors in its decision:
- The nature and profit model of the GenAI service: The defendant directly obtains economic benefits from the creative services provided by the platform.
- The popularity of the copyrighted work and the obviousness of the alleged infringement: Ultraman works are quite well-known. When browsing the platform homepage and specific categories, there are multiple infringing pictures, and the LoRA model cover or sample picture directly displays the infringing pictures, which is a relatively obvious infringement.
- The infringement’s consequences: The pictures and LoRA models generated and published by users can be repeatedly used by other users. The trend of causing the spread of infringement consequences is already quite obvious, and the defendant should have foreseen the possibility of infringement.
- Whether reasonable measures have been taken to prevent infringement: The defendant stated in the platform user service agreement that it would not review the content uploaded and published by users. The defendant should have known that network users used its services to infringe upon the right of information network dissemination but did not take necessary prevention measures.
The Hangzhou Internet Court concluded that the defendant “failed to fulfil its duty of reasonable care” and was subjectively at fault, constituting “aiding and abetting infringement”. However, it held that there was not enough evidence to prove that the defendant and the users jointly provided infringing works, nor that it “directly” implemented actions protected by information network dissemination rights.
This decision follows several others handed down in China last year, including by the Changshu Court in October which found that copyright law does not protect “ideas or concepts”. In Lin Chen v. Hangzhou Gaosi Membrane Technology, the plaintiff used AI to create a visual artwork called “With Heart”. The court found that the “With Heart” image was distinctly original in its composition and the arrangement of elements such as the cityscape, water, buildings and reflections – it thereby qualified as a visual artwork under copyright law and merited protection. However, copyright protection was only extended to this 2D work and not the 3D installation. So this “similar work” did not constitute copyright infringement, but the unauthorised use of the plaintiff’s image for online promotion, which was found to be nearly identical to the original, did infringe upon the plaintiff’s right to distribute the work online.
In another decision by the Beijing Internet Court last April, the court held for the first time that an AI-generated voice can be linked to an individual and can therefore be copyright-protected. A natural voice – distinguishable by tone, pitch and frequency – is unique and identifiable, thereby enabling an audience to associate it with a specific person, it stated. So, if it can be linked to a person based on these circumstances, then the individual’s personality rights extend to the AI-generated product too.