01
Factual background & dispute
- Shanghai Character License Administrative Co., Ltd. held an exclusive licence in China to the reproduction, adaptation and information-network dissemination rights in the Ultraman artworks and had independent enforcement authority.
- The defendant's Tab website connected to a third-party AI image service through an API. Prompts such as 'generate an Ultraman Dyna' produced images partly or wholly substantially similar to protected Ultraman characters.
- The defendant argued that a third party supplied the image function and that it had not trained the model. The court found that the website lacked a complaint mechanism, intellectual-property risk notices and conspicuous AI-content labels.
02
Core issues & judicial focus
- Whether the generated images infringed the reproduction, adaptation and information-network dissemination rights
- Whether an operator using a third-party model through an API qualifies as a generative AI service provider
- What technical prevention standard applies to cessation of infringement and whether training data must be deleted
- How complaint handling, risk notices and AI-content labelling affect damages liability
03
Judicial finding & holding
- The court found that some outputs reproduced original elements of Ultraman and that others retained those elements while adding new features, infringing the reproduction and adaptation rights. It did not separately repeat the analysis of the information-network dissemination right.
- Because the defendant provided the AI image service to users through an API, the court treated it as a generative AI service provider and required technical measures preventing substantially similar outputs during ordinary use of Ultraman-related prompts.
- The defendant had not trained the model, so the court rejected the request to delete training data. It awarded RMB 10,000 including reasonable expenses.
04
Practical risk implications
01Product operators remain responsible for statutory duties when they deliver generative AI functions through third-party models or APIs.
02Highly recognisable intellectual property requires keyword and semantic controls, similarity testing, complaint handling and continued regression testing.
03User terms should define intellectual-property limits, generated content should carry required labels, and testing and remediation should be documented.