First AI-Generated Picture Copyright Case in the PRC

On 27 November 2023, the Beijing Internet Court handed down a first-instance civil case judgment. The case, centring around the copyrightability of a two-dimensional picture generated by artificial intelligence (AI) in response to human prompts, was reported as the “first AI-generated picture copyright case” in the People’s Republic of China (the PRC).

The plaintiff, an individual, employed open-source software, Stable Diffusion, to create a portrait of a woman. For the creation of this picture, the plaintiff put in a series of positive prompts to direct the AI in incorporating desired features into the picture, and also a series of negative prompts to specify elements to be avoided. The picture so generated was posted by the plaintiff on RED, a prominent social media platform focused on youth lifestyle in the PRC. The defendant, also an individual, utilised the picture without the plaintiff’s authorisation in an article published on another platform. In so doing, the defendant also removed the watermarked user number in the picture that was assigned by RED to the plaintiff.

The core issue in this case is whether this AI-generated picture, created with the prompts inputted by a human being, constitutes a “work” eligible for protection under the PRC Copyright Law. In analysing this issue, the Beijing Internet Court delved into two primary considerations:

As a background note, the PRC Copyright Law defines copyrightable works as “original intellectual achievements in the fields of literature, art, and science that can be presented in a certain form”.

The court addressed the issue of intellectual achievement first, providing an affirmative answer. In the court’s language, the test should be whether the picture had “intellectual contribution” instilled by a human being. The court indicated that during the creation process of the picture, the plaintiff injected their intellect, evidenced by, on a non-exhaustive basis, actions such as designing the woman’s appearance, selecting prompts, organising such prompts sequentially, setting parameters, and refining the image by adjusting prompts to choose from multiple results. Such actions, in the court’s view, constitute intellectual contribution that elevates the picture to the status of an intellectual achievement. In this creative process, the court considered the AI, “at its current stage”, still merely a tool for creation, because the AI “does not possess free will.”

The court next addressed the issue of originality, emphasising the factor of individualised expression. In the court’s view, when a human being utilises AI to generate pictures, the more distinctive and specific the requests, the easier it is to establish individualised expression for the purpose of originality. In this case, the plaintiff used dozens of prompts for the creation, demonstrating a fair level of specificity.

Notably, despite their abundance, all the prompts in Chinese would encompass only about half a page (slightly longer in English) with an 11-point font. This touches on a relevant question of the minimum level of prompts required to confer “originality” upon the generated results. Considering the traditionally low threshold for copyrightability, such as in the case of photography, it is expected that the threshold for an AI-generated picture will also be relatively modest. Based on copyright principles, it is conceivable that overly simplistic prompts could render the results ineligible for copyright protection. For example, a picture generated by AI with a single word or an excessively generic phrase like “blue sky” may not attain copyrightability. However, in real-world scenarios addressing this issue (ie, copyrightability under simple prompts) in court would likely be a situation of case-by-case examination and discussion.

After acknowledging the copyrightability of the picture and establishing the plaintiff’s authorship, the court turned to the issue of infringement on the author’s right to attribution. The court found that the defendant, in using the picture in question, had removed the watermarked user number that RED assigned to the plaintiff. Emphasising that this number exclusively corresponded to the plaintiff, the court highlighted the watermark’s role in signifying the plaintiff’s authorship. Furthermore, the plaintiff expressly stated their choice to use this number as the credit for the work. Based on the above, the court held that the unauthorised removal of the watermarked user number constituted an infringement on the plaintiff’s right to attribution under the PRC Copyright Law.

The Beijing Internet Court is a specialised forum established in 2018 dedicated to adjudicating diverse internet-related cases. As Beijing courts in general are known for their expertise with respect to cases in the intellectual property law areas, one can expect that the perspectives, analyses and holdings established in this case will likely be considered and referred to in future legal proceedings.