Facts
The plaintiff, a designer, used Midjourney and similar AI drawing tools in August 2023 to generate images of a transparent, butterfly-shaped children's chair and posted them on Xiaohongshu together with the prompts used. After an offer of cooperation was turned down, the defendant Zhu drew on the published prompts to generate new butterfly-chair designs with similar tools, which a company then manufactured and sold from 2024. The plaintiff claimed copyright infringement and unfair competition, seeking an end to sales, destruction of stock and moulds and CNY 200,000 in damages. Asked by the court to re-enact the process at the hearing, the plaintiff could not reproduce the same images.
Question
The first question was whether AI-generated images are works, meaning original intellectual creations, under China's Copyright Law. That turned on how a user proves an original human contribution to the generation process: whether revising prompts, adjusting parameters, and selecting and refining outputs is enough, and what records must show it. The court also had to decide whether published prompts are themselves protected, and whether the defendants' products, promotional images and packaging were substantially similar to the plaintiff's images.
Decision
The court held that content generated mainly and automatically by AI cannot be presumed to be a work. A user claiming copyright must prove an original intellectual contribution beyond supplying an initial idea or simple instructions. The plaintiff produced no original records of the prompts, parameter changes and selections behind the three images and could not reproduce them, so they were not works. Prompts such as ‘jelly texture’ and ‘butterfly shape’ are ideas rather than protected expression, and using published prompts was not infringement. The products and images also differed markedly in concrete expression, so there was no substantial similarity. With no commercial use of the images shown, there was no unfair competition either, and all claims were dismissed.
Why it matters
The court's release describes the case as China's first holding that an AI text-to-image output is not a work, and the Suzhou Intermediate People's Court listed it among its typical IP cases of 2025. The Beijing Internet Court's 2023 ruling in Li v. Liu had protected an image because the user's detailed input was shown. The ruling draws its line on proof as much as principle: protection depends on concrete contributions, such as several rounds of prompt revision and parameter setting, and on keeping records of the generation process. Basic descriptive prompts were treated as unprotected ideas.
Related stages
According to Legal Daily, the Suzhou Intermediate People's Court treated the plaintiff's appeal as withdrawn after the appeal fee went unpaid, and the first-instance judgment took effect; the court's own release says the plaintiff appealed and then withdrew the appeal. The court sources give no judgment date. IAM reports 19 March 2025, which no court or Chinese source opened for this entry confirms, so only the month is given. In a case reported in September 2026, the Changsha Kaifu District Court likewise held that an image generated from prompts setting only subject, colour and style was not a work (Gmw.cn, 22 September 2026).