Shanghai Court Issues First Instance Judgment on AI Model Infringement

On November 3, 2025, the Jinshan District People's Court of Shanghai (the “Court”) delivered its first-instance judgment in Shanghai's first AI large model copyright infringement case. The Court determined that the defendant AI image generation platform did not infringe upon the plaintiff's rights, as it had no subjective fault and had objectively fulfilled its obligations to “take necessary measures” and “notify others”, and therefore it did not constitute infringement. However, the platform user, an individual defendant named Li, was found to have infringed the plaintiff's reproduction rights and information network dissemination rights for the character image of Medusa in the “Battle Through the Heavens” animated series. 

Case Introduction

The plaintiff company in this case holds the copyright to the character image of “美杜莎 (Medusa)” from the animated series adapted from the novel “Battle Through the Heavens”; whereas the two defendants are an AI image generation platform (“the defendant platform”) -- reportedly “aggregating numerous AI image generation LORA models across various types and themes, leveraging large models and multiple LORA models to provide users with AI online image generation services”-- and an individual defendant named Li, who is a user of the defendant platform.

It is reported that the defendant Li extracted over twenty images of Medusa from the plaintiff's “Battle Through the Heavens” animated series, compiled them into a Medusa image pack, and then “fed” this material to the “training LoRA function” of the defendant platform to generate two Medusa LoRA models. The defendant Li then published the involved Medusa LoRA models, which had undergone automated review by the defendant platform, on his account. When other users employed these involved LoRA models, they could generate various images identical or substantially similar to the plaintiff’s Medusa appearance simply by inputting different prompt words.

Accordingly, the plaintiff claimed that since its “Medusa” constitutes a commodity name with certain influence and the Medusa LoRA model in question can specifically generate Medusa character images, the defendant Li constituted unfair competition and infringed upon the plaintiff's rights to reproduce, adapt, and disseminate the Medusa image via information networks. Furthermore, the defendant's platform not only hosted numerous LoRA models infringing the plaintiff's Medusa character image, but its “Anime Zone” also contained a large number of LoRA models suspected of infringing other character images from the novel Battle Through the Heavens. The defendant platform had neglected to address these infringements and failed to fulfill its platform responsibilities. Therefore, the plaintiff brought this action and requests the court to order the two defendants to 1) immediately cease copyright infringement and unfair competition activities; 2) issue a public statement to eliminate the adverse effects of their infringing acts on the plaintiff; and 3) compensate the plaintiff by the defendant platform for economic losses and reasonable expenses incurred in protecting its rights totaling RMB 2 million, while defendant Li compensates the plaintiff for economic losses and reasonable expenses totaling RMB 200,000. Both defendants shall bear joint and several liability for compensation up to RMB 150,000.

The defendant platform argued that “Medusa” primarily refers to the Gorgon from Greek mythology and is not an original creation of the plaintiff, and therefore should not be protected as a commodity name with certain influence. Furthermore, since the LoRA model itself and its “Train LoRA” feature constitute technology advancing AI development, they inherently lack infringing attributes. The defendant platform had promptly taken necessary measures, such as removing the content and notifying overseas AI platforms sharing the data, thereby fulfilling its notice-and-takedown obligations. Consequently, the defendant platform contended that it did not commit infringement. On the other hand, the defendant Li recognized his mistake in the Court and had completely ceased all infringing activities.Upon trial, the Court considered that the term “Medusa” is most widely recognized as the Greek mythological Gorgon with serpent hair, rather than an original creation by the plaintiff. Given its rich connotations, numerous characters in comics, novels, and games, song titles, plant varieties, and so on have been named “Medusa.” Therefore, the Court does not support the plaintiff's claim that ‘Medusa’ constitutes a “commodity name with certain influence” for protection.

In terms of the copyright infringement, the Court considered that “the defendant Li, for commercial purposes, reproduced the original expression of the prior work during the material extraction phase and the training, release, and use of the LoRA model. By making the ‘Medusa’ image collection and short videos available to the public via the internet, he infringed upon the plaintiff's reproduction rights and information network dissemination rights for the ‘Medusa’ work. In the absence of evidence demonstrating substantial intellectual contribution by a natural person, images directly generated by the generative AI do not constitute artistic works under the Copyright Law. In this case, the Medusa LoRA model and AI-generated images did not involve substantial intellectual input from the defendant Li, thus failing to qualify as works. Consequently, the defendant Li did not infringe upon adaptation rights.”

Regarding the liability of the defendant platform, the Court considered that “based on the characteristics of the services and technologies provided by the defendant platform, it did not directly participate in the material extraction, training, release, or use of LoRA models. The LoRA model technology itself and the functionality for training LoRA models constitute neutral technologies that facilitate the development of generative artificial intelligence. Therefore, the defendant company qualifies as an online service provider. Whether the platform infringed in this case should be determined comprehensively by considering both subjective and objective elements. The platform fulfilled its reasonable duty to inform users, established complaint reporting and content review mechanisms, promptly removed all Medusa LoRA models upon receiving the complaint, updated screening keywords in its review system, and notified overseas AI platforms after receiving the plaintiff's notice. Subjectively, it bears no fault; objectively, it fulfilled its obligations to “take necessary measures” and “forward notifications.” It should not be deemed liable for infringement.”

In conclusion, the Court ordered in its first instance judgment that “1) the defendant Li cease infringing upon the plaintiff's reproduction rights and information network dissemination rights for the character image of Meidu Sha in the “Battle Through the Heavens” animated series; 2) the defendant Li compensate the plaintiff for economic losses of 30,000 yuan and reasonable expenses incurred in protecting its rights of 20,000 yuan within ten days from the date this judgment takes effect; 3) the plaintiff's other claims are dismissed.”
Brief Comment
This case primarily concerns how to determine platform liability versus user liability in cases involving copyright infringement by large AI models.
In accordance with Article 10 of the PRC Copyright Law “Copyright shall include the following personal rights and property rights: … (5) the right of reproduction, that is, the right to produce one or more copies of the work by means of printing, Xeroxing, rubbing, sound recording, video recording, duplicating, or re-shooting, etc.; … (12) the right of information network dissemination, that is, the right to provide the public with works by wired or wireless means, so as to make the public able to respectively obtain the works at the individually selected time and place; … (14) the right of adaptation, that is, the right to modify a work for the purpose of creating a new work of original creation; …”

In this case, the defendant Li extracted the “Medusa” image from the plaintiff's animated work, created an image pack, utilized the defendant platform's LoRA technology to fine-tune a proprietary model, and publicly released it. These actions enabled the public to generate various images identical or substantially similar to the original work through prompt words. As such actions satisfied the “substantial similarity + access” rule, the Court determined that the defendant Li had infringed upon the plaintiff's reproduction rights and information network dissemination rights for the ‘Medusa’ character in the animated work “Battle Through the Heavens,” and ordered him to pay 30,000 yuan in economic losses and 20,000 yuan in legal expenses. This judgment sends a clear signal to the AI platform users that training AI models by using others' original character designs and then sharing the results is no longer a “gray area.” Individual users may also bear direct liability for infringement.

It is worth noting that the court did not recognize that the defendant Li's actions infringed upon the plaintiff's adaptation rights to the relevant image. This is because the Court considered that the images directly generated by the generative AI in this case, lacking evidence of substantial human creative input, did not constitute artistic works under the Copyright Law. Article 3 thereof stipulates that “Works mentioned in this Law shall include works of literature, art, natural science, social science, engineering technology, ...” Consequently, this case offers valuable insights into whether AI-generated content necessarily constitutes a work under the Copyright Law and whether utilizing copyrighted materials for AI creation and model training infringes upon the copyright holder's adaptation rights.Regarding the issue of the defendant platform's liability for the infringement, the Court considered the following factors: the defendant platform acted solely as a neutral technology and functionality provider in this case, did not participate in the extraction of source materials or model training, and promptly removed the infringing model and updated keyword filters upon receiving the complaint, thereby fulfilling its “notice-and-takedown” obligations. With no subjective fault, the Court ultimately determined that the defendant platform bears no joint liability.

Among them, the primary provisions regarding the “notice-and-takedown” obligations of the defendant platform as an online service provider are set forth in Article 1195 of the Civil Code, which provides that “Where a network user commits a tortious act through using the network service, the right holder is entitled to notify the network service provider to take such necessary measures as deletion, block, or disconnection. The notice shall include the preliminary evidence establishing the tort and the real identity information of the right holder. After receiving the notice, the network service provider shall timely forward the notice to the relevant network user and take necessary measures based on the preliminary evidence establishing the tort and the type of service complained about. Where it fails to take necessary measures in time, it shall assume joint and several liability for the aggravated part of the damage with the network user. The right holder who causes damage to the network user or network service provider due to erroneous notification shall bear tort liability, unless otherwise provided by law.” Further, Article 1197 of the Civil Code provides that “A network service provider, who knows or should have known that a network user has infringed upon the civil-law rights and interests of another person by using its network services but fails to take necessary measures, shall assume joint and several liability with the network user.”

It can be seen that the said provisions establish a compliance baseline for all types of online service providers, including the generative AI platforms. Specifically, such platforms must establish reasonable content review, complaint response, and risk warning mechanisms and strictly enforce these mechanisms to achieve a “safe harbor” between technological innovation and the protection of rights such as copyright—as exemplified by the judgment in this case.

In conclusion, against the backdrop of increasingly widespread generative AI applications, striking a balance between encouraging technological innovation and safeguarding the legitimate rights of original creators has become an increasingly prominent issue. This case offers valuable insights into defining “user responsibility” and “platform responsibility.” For platform users, this case serves as a cautionary reminder that they bear legal responsibility for feeding copyrighted content to large AI models and for the resulting consequences. For AI platforms, it provides guidance that they must establish and rigorously enforce reasonable content review, complaint response, and risk disclosure mechanisms in accordance with relevant laws to avoid unnecessary legal risks.