Facts
The plaintiff, Shanghai Xinchuanghua, held an exclusive licence in China from Japan’s Tsuburaya Productions for the Ultraman works. The defendant ran a website called Tab offering AI chat and image generation; the drawing function was open only to users who paid for membership and “computing power”. The plaintiff found that prompts such as “generate an Ultraman” produced images substantially similar to Ultraman characters. The defendant had not trained the model itself and provided the service through a third party’s model. The plaintiff sued for RMB 300,000 in damages, an injunction and the removal of Ultraman material from the training data.
Question
The court had to decide whether a provider of a generative AI service is liable for copyright infringement when user prompts produce images substantially similar to a protected work. That required applying the tests of access and substantial similarity, and deciding whether the outputs infringed the right of reproduction, the right of adaptation, or both. The court also considered the duties of care owed by a service provider under the Interim Measures for the Management of Generative AI Services, and whether a provider that had not trained the model could be ordered to delete training data.
Decision
On 8 February 2024 the court held that the defendant had infringed the plaintiff’s rights of reproduction and adaptation. Access was inferred from the wide availability of Ultraman on major video platforms. Some images reproduced the original expression in part or in full, while others kept the characters’ essential features and added new elements. Providers, the court said, should meet reasonable and affordable duties of care, such as complaint mechanisms, risk warnings and prominent labelling of generated content. The defendant was ordered to take technical measures against generating substantially similar Ultraman images in normal use and to pay RMB 10,000, including reasonable costs. The training-data deletion request failed because the defendant had not trained the model.
Why it matters
The judgment is reported as one of the first in China to hold a generative AI service provider liable for infringing output, and it has taken effect. The court stressed that the industry is at an early stage and that providers should not be overburdened, while still expecting them to meet reasonable duties of care. It also distinguished the developer that trains a model from the platform that offers it to users. The lawfulness of the training itself was not examined in this case.