Changsha Court Examines AI-Generated Images in Three Layers: Generic Prompts Plus a Simple Photoshop Edit Do Not Make a Copyright Work

A copyright case decided by Changsha's Kaifu District People's Court has become the focus of professional discussion after being reported in the media: the court divided the process of producing AI-generated content into three layers, front-end conception, generation control and back-end processing, examined originality layer by layer, and held that the image in dispute is not a work within the meaning of the Copyright Law, dismissing all of the plaintiff's claims. According to a report by Xiaoxiang Morning Post on 22 September, a third party generated the image on 8 May 2025 with the Jimeng AI image model by entering the prompt 'Dragon Boat Festival sachet, colourful, Chinese style, with the Chinese characters for Dragon Boat Festival'.

Duplicated characters at the bottom of the image were then removed in Photoshop; on 3 June the economic rights were assigned to the plaintiff and on 18 June a copyright registration was completed; a Changsha kindergarten used the image on its WeChat official account and the plaintiff sued for infringement of its right of communication over information networks.

The court held that whether an AI-generated image is a work cannot be decided by the generating tool alone but by whether it reflects a natural person's original intellectual input; at the generation control layer the key to originality lies in the author's selection and arrangement of the elements of expression, examined from three angles: whether the controlling act is directed at expression itself, whether there is a correspondence between the prompt and the picture, and whether the user's choices play the dominant role.

Here the prompt, beyond naming the object, contained only general requirements of subject, colour and style such as 'colourful' and 'Chinese style', with no individualised choices about the sachet's shape, the composition, the arrangement of elements or light and shade, and resolution and aspect ratio were ordinary technical settings; the back-end processing left the picture essentially unchanged apart from removing duplicated characters, a technical correction that does not amount to original expression.

Neither party appealed and the judgment has taken effect; the presiding judge added that a few prompts do not by themselves confer copyright, and that users seeking protection should make individualised choices about the picture's expression and keep records of prompts, parameters, selection and modification.

This is not a new judgment today but an effective decision now under concentrated professional discussion: it does not contradict the Beijing Internet Court's 2023 finding that an AI image produced through many rounds of prompt and parameter adjustment was protected, and it continues the line taken by the Zhangjiagang court requiring records of the creative process, placing the focus on whether human choices directly controlled the specific expression.

For companies, whether AI output can be a protected work depends on provable human control during generation, not on a registration certificate or on the tool itself.

J&C recommends that companies using AI for marketing visuals or product images work on the basis of provable control from the outset: iterate with prompts directed at specific expression such as composition, shape, arrangement and lighting, keep every round of prompts, parameters, candidate images and selection reasons, and carry out substantive post-processing; a copyright registration is only prima facie evidence and cannot replace process records; and do not read such judgments as a licence to use other people's AI images freely, since source and authorisation should still be checked.

Source: m.gmw.cn