On 10 September 2025, the Beijing Internet Court of the People’s Republic of China (“Court”) released eight typical cases involving artificial intelligence (“AI”) at a press conference on the adjudication of AI-related disputes. Through these cases, the Court clarified several judicial rules for the review of AI-related cases, such as determination of the legal nature and ownership rights of AI-generated text-to-image content, extensibility of natural persons’ voice rights to AI-generated voices, infringement arising from unauthorized use of AI-synthesized celebrity voices for commercial endorsements, violation of portrait rights through unauthorized “AI face-swapping” of videos containing personal likenesses, obligations of online platforms to provide reasonable explanations when using algorithmic tools to detect AI-generated content, infringement resulting from the use of AI software to maliciously distort personal likenesses, eligibility of original virtual human images for protection as works of fine art, and infringement constituted by creating AI-generated likenesses of natural persons without consent.
Typical Case I: A dispute involving infringement of authorship right and information network dissemination right between Li Moumou and Liu Moumou – This case establishes that content generated by natural persons using AI, if satisfying the definition of a work, shall be recognized as a work protected under the Copyright Law of the People’s Republic of China (“Copyright Law”). Ownership of copyright may be determined based on the original intellectual contributions of various participants, such as developers, users, and owners, to the AI-generated content.
On 24 February 2023, the plaintiff, Li Moumou, utilized the generative AI model Stable Diffusion to produce a photographic-style close-up portrait of a young woman under nightfall lighting conditions. Through selecting the model, inputting prompt and negative prompt words, and setting generation parameters, the plaintiff generated an image and published it on the Xiaohongshu platform on 26 February 2023. On 2 March 2023, the defendant Liu, Moumou, published an article on his Baijiahao account, using the image as an illustration. The defendant cropped out the plaintiff’s signature watermark and did not indicate the specific source of the image. The plaintiff thus filed a lawsuit, requesting a public apology and compensation for economic losses.
During court inspection, it was demonstrated that altering individual prompt words or parameters resulted in the generation of different images by the AI model. The court held that the image with a specific form of expression, in terms of its appearance, possesses indistinguishable from ordinary photographs or paintings and belongs to the field of art. The image was generated by the plaintiff using generative AI technology. From the conception to the final selection of the image, the plaintiff made substantial intellectual investments, thus rendering the image an “intellectual achievement”. The image exhibited identifiable differences from prior works, reflecting the plaintiff’s selections and arrangements. The process of adjustment and modification also demonstrated the plaintiff’s aesthetic choices and personal judgment. In the absence of evidence to the contrary, the image was independently completed by the plaintiff and embodied his personalized expression, thereby satisfying the requirement of “originality” for copyright protection. As a two-dimensional artistic work composed of lines and colors with aesthetic significance, the image qualifies as a work of fine art protected under the Copyright Law. Regarding copyright ownership, the Copyright Law stipulates that authors are limited to natural persons, legal persons, or unincorporated organizations. Therefore, the AI model itself cannot be considered an author under the Copyright Law. The plaintiff, as the natural person who configured the AI model according to specific needs and ultimately selected the image, is the author of the image entitled to copyright protection. The image was directly generated based on the plaintiff’s intellectual input and reflected their personalized expression. The defendant’s unauthorized use of the image as an illustration, making it accessible to the public at chosen times and locations, infringed upon the plaintiff’s information network dissemination right. Moreover, the removal of the plaintiff’s signature watermark violated the plaintiff’s authorship right. The defendant was thus liable for copyright infringement. The court ordered the defendant to issue a public apology and provide compensation for economic losses. Neither party appealed the judgment, which has now taken effect.
Typical Case II: A personal right infringement dispute between Yin Moumou and Mou Smart Technology Company et al. – This case establishes that a natural person’s right of voice extends to AI-generated voices and clarifies the criteria for determining whether voices processed by AI technology are protected under the right of voice.
The plaintiff, Yin Moumou, a voice actor, discovered that works generated using his voice recordings were widely disseminated on multiple well-known apps. Through voice screening and tracing, it was determined that the voices in these works originated from a text-to-speech product available on a platform operated by the first defendant, Mou Smart Technology Company. The plaintiff had previously been commissioned by the second defendant, Mou Cultural Media Company, to produce sound recordings, of which the second defendant is the copyright holder. The second defendant subsequently provided the plaintiff’s recordings to the third defendant, Mou Software Company. Using only one recording by the plaintiff as source material, the third defendant processed it through AI technology to develop the text-to-speech product in this case, which was then offered for sale on a cloud service platform operated by the fourth defendant, Mou Internet Technology Company. The first defendant entered into an online service agreement with the fifth defendant, Mou Technology Development Company, which placed an order with the third defendant that included the disputed text-to-speech product. The first defendant employed an application programming interface to directly access and generate the text-to-speech product on its platform without additional technical processing. The plaintiff claimed that the defendants’ actions seriously infringed his voice rights, and requested that the first and third defendants cease infringement and issue a public apology, and that all five defendants compensate the plaintiff for economic and moral damages.
The court held that a natural person’s voice, distinguished by its vocal print, timbre, and frequency, is unique, distinctive, and stable, capable of evoking associations or emotional responses related to that individual, thereby reflecting personal behavior and identity. Recognizability of a natural person’s voice means that upon repeated or prolonged exposure, the voice can identify a specific natural person. An AI-synthesized voice is considered recognizable if it enables the public or relevant audiences to associate the voice, based on its timbre, tone, and pronunciation style, with that natural person. In this case, as the third defendant developed the text-to-speech product exclusively using the plaintiff’s voice, and given that during in-court inspection the AI-generated voice was highly consistent with the plaintiff’s timbre, tone, and pronunciation style, it was capable of evoking associations with the plaintiff and identifying the plaintiff’s identity. While the second defendant held copyrights to the sound recordings, this did not include the right to authorize others to use the plaintiff’s voice for AI applications. The agreement between the second and third defendants, which authorized the AI use of the plaintiff’s voice without their informed consent, lacked legal basis. Thus, the defense of lawful authorization raised by the second and third defendants was rejected. The unauthorized AI use of the plaintiff’s voice by the second and third defendants constituted infringement of the plaintiff’s right of voice, resulting in harm for which they bear legal liability. The first, fourth, and fifth defendants were found subjectively free of fault and not liable for damages. Considering factors such as the nature of the infringement, market value of comparable products, and product usage metrics, damages were determined at the court’s discretion. The court ordered the first and third defendants to provide a written apology to the plaintiff, and the second and third defendants to compensate the plaintiff for economic losses. Neither party appealed the judgment, which has now taken effect.
Typical Case III: An internet infringement dispute between Li Moumou and Mou Cultural Media Co., Ltd. – This case establishes that unauthorized use of AI-synthesized celebrity voices for “live commerce” constitutes infringement. It clarifies that voices synthesized using AI deep synthesis technology fall within the protection scope of a natural person’s right of voice if they enable the public or relevant audiences to identify the specific natural person based on vocal characteristics such as timbre, tone, and pronunciation style. Additionally, in principal-agent relationships where merchants commission video promoters for sales, the merchant—as the commissioning party and actual beneficiary—bears a reasonable duty to review promotional content published by its associated video promoter. Failure to fulfill this duty results in joint liability with the video promoter.
The plaintiff, Li Moumou, possesses significant recognition and social influence in the fields of education and parenting. In 2024, the plaintiff discovered that the defendant, Mou Cultural Media Co., Ltd., utilized the plaintiff’s public speeches and instructional videos, paired with an AI-synthesized voice highly similar to the plaintiff’s, to promote multiple family education books sold through the defendant’s online store on a digital platform. The plaintiff alleged that the defendant’s unauthorized use of his portrait and AI-synthesized voice in promotional materials created a strong association between the plaintiff’s personal image and the defendant’s commercial promotions, misleading consumers into believing the plaintiff endorsed or recommended the books. By leveraging the plaintiff’s personal image, professional background, and social influence to attract attention and increase sales opportunities, the defendant infringed upon the plaintiff’s right of portrait and right of voice. As the book seller, the defendant maintained a principal-agent relationship with the video promoter (a live commerce streamer) to jointly conduct sales activities. The defendant possessed both the duty and capacity to review the promoter’s videos and was thus liable for the publication of the infringing content, including obligations to apologize and compensate for losses.
The court held that the video incorporated the plaintiff’s portrait and an AI-synthesized voice highly consistent with the plaintiff’s natural voice in timbre, tone, and pronunciation style. Given the plaintiff’s prominence in education and parenting, the video’s promotion of family education books made it likely for viewers to associate its content with the plaintiff. It was determined that relevant audiences could directly correlate the AI-synthesized voice with the plaintiff. Therefore, the voice fell within the protection scope of the plaintiff’s right of voice. The substantial use of the plaintiff’s portrait and synthesized voice without authorization constituted infringement of the plaintiff’s right of portrait and right of voice. The defendant and the video promoter had established a commissioned promotion relationship under platform rules and service agreements to jointly promote the defendant’s books and generate revenue. Based on platform rules and management authority, the defendant had the ability to review and manage the video content. Given the prominent use of the plaintiff’s portrait and synthesized voice, the defendant should have foreseen potential infringement risks and was obligated to reasonably verify whether the video had obtained the plaintiff’s authorization. Evidence indicated that the defendant failed to exercise due diligence in review. Thus, the defendant was held jointly liable with the video promoter for the publication of the infringing video. The court ordered the defendant to issue a formal apology to the plaintiff and compensate for economic losses and reasonable litigation costs. The plaintiff’s other claims were dismissed. Neither party appealed the judgment, which has now taken effect.
Typical Case IV: An internet infringement dispute between Liao Mou and Mou Science and Technology Culture Co., Ltd. – This case addresses the emerging business model of “AI face-swapping” and accurately distinguishes between portrait rights, personal information rights, and legally protected interests derived from creative labor in the application of generative synthetic AI. It establishes that unauthorized “AI face-swapping” of videos containing another person’s portrait constitutes infringement of personal information rights.
The plaintiff, Liao Mou, is a short-video blogger specializing in classical Chinese-style content with a substantial following across multiple platforms. Without authorization, the defendant, Mou Science and Technology Culture Co., Ltd., used a series of videos featuring the plaintiff to create face-swapping templates, which were uploaded to its software and offered to users for paid access, thereby generating profits. The plaintiff alleged that the defendant’s actions infringed upon his rights of portrait and personal information, demanding a written apology and compensation for economic and moral damages. The defendant contended that the videos published on its platform had lawful sources, that the facial features in the templates were not those of the plaintiff, and thus no portrait right infringement had occurred. Additionally, the defendant argued that the face-swapping technology was provided by a third party, and therefore it had not processed the plaintiff’s personal information.
The court found that the face-swapping template videos exhibited identical characteristics to the plaintiff’s original works in terms of makeup, hairstyle, clothing, movements, lighting, and scene transitions, but the facial features in the templates were entirely different and did not belong to the plaintiff. The defendant’s software utilized third-party services to implement the face-swapping function, and users could unlock all features by paying a membership fee. The court held that the determinative factor for portrait right infringement is recognizability. Recognizability emphasizes that the essence of a portrait lies in its ability to identify a specific natural person. While the scope of a portrait primarily centers on facial features, it may also include distinctive body parts, voices, or highly identifiable actions that correspond to a specific natural person. In this case, although the defendant used the plaintiff’s videos to create templates, it did not utilize the plaintiff’s portrait but instead replaced the facial features through technological means. Elements such as makeup, hairstyle, clothing, lighting, and scene transitions retained in the templates are not inseparable from a specific natural person and differ from inherent personal attributes. The public would identify the subject in the altered videos as a third party rather than the plaintiff. Moreover, the defendant’s provision of these templates to users did not involve defamation, distortion, or forgery of the plaintiff’s portrait. Thus, the defendant’s actions did not constitute infringement of the plaintiff’s right of portrait. However, the defendant collected videos containing the plaintiff’s facial information and replaced the plaintiff’s face with faces from other provided photos. This synthesis process required algorithmic integration of features from new static images with partial facial characteristics and expressions from the original videos. This procedure involved the collection, use, and analysis of the plaintiff’s personal information, constituting processing of personal information. The defendant’s processing of this information without the plaintiff’s consent infringed upon the plaintiff’s personal information rights. Regarding the unauthorized use of videos created by others, which may infringe upon rights derived from creative labor, such claims should be asserted by the relevant rights holders. The court ordered the defendant to issue a written apology to the plaintiff and compensate for moral damages and emotional distress. The plaintiff’s other claims were dismissed. Neither party appealed the judgment, which has now taken effect.
Typical Case V: A network service contract dispute between Tang Moumou and Mou Technology Co., Ltd. – This case establishes that online content platforms utilizing algorithmic tools to detect AI-generated content bear liability for breach of contract if they fail to fulfill reasonable and appropriate explanation obligations. While affirming the positive role of platforms in reviewing and managing AI-generated content through algorithmic tools to fulfill their information management responsibilities, the case clarifies that platforms must provide adequate explanations for automated decision-making results in scenarios involving real-time creative text.
The plaintiff, Tang Moumou, published a text of approximately 200 characters on a platform operated by the defendant, Mou Technology Co., Ltd, stating: “Work cannot truly earn you much money, but it can open new perspectives... If you are interested in learning to drive and plan to do so in the future, complete it during your leisure holiday... After starting work, there will be little uninterrupted time to obtain a driver’s license”. The defendant’s platform flagged the content as violating its policy for “containing AI-generated content without labeling”, hid the content, and imposed a one-day account suspension on the user. The plaintiff’s subsequent appeal was unsuccessful. The plaintiff claimed that the content was human-generated and that the defendant’s actions constituted a breach of contract, requesting the court to revoke the penalty, restore the content, and delete the violation record from the backend system.
The court held that users employing AI tools for content creation must label such content honestly and in accordance with the principle of good faith. The defendant had issued a community announcement requiring creators to actively label AI-generated content (AIGC) upon publication. The announcement stipulated that content without explicit labeling would be subject to circulation restrictions and retroactive labeling by the platform, and that unlabeled use of AIGC constituted a violation. As a registered user, the plaintiff was bound by this announcement, which formed part of the platform’s service agreement. The defendant was entitled to review and process user content to determine whether it was AI-generated. Generally, the plaintiff should provide preliminary evidence of human authorship, such as drafts, originals, source files, or source data. However, in this case, the plaintiff’s content was created in real time, making it objectively impossible to provide such evidence. The defendant’s conclusion that the content was AI-generated was based solely on algorithmic tool results. As the controller of both the algorithmic tool and the decision-making process, the defendant bore the responsibility to provide reasonable proof or explanation for its determination. Although the defendant submitted algorithmic filing information, it failed to demonstrate its relevance to the disputed content. The defendant did not adequately explain the basis or results of its algorithmic decision, nor did it justify its determination that the content was AI-generated. Consequently, the defendant was liable for breach of contract due to its lack of factual basis in penalizing the plaintiff’s account. The court further noted that the defendant’s manual review standards, which relied on subjective assessments of “distinct human emotional characteristics”, lacked scientific basis, persuasiveness, and credibility. The court ordered the defendant to restore the folded content and delete the relevant backend records. The defendant appealed the first-instance judgment but later withdrew the appeal, rendering the judgment effective.
Typical Case VI: An internet infringement dispute between Cheng Mou and Sun Mou – This case establishes that unauthorized use of AI software to mock or distort another person’s portrait constitutes infringement of personality rights. It emphasizes that users of generative AI technology must comply with laws, respect social morality and ethical principles, and refrain from endangering the legitimate rights and interests of others.
The plaintiff, Cheng Mou, and the defendant, Sun Mou, were members of the same photography discussion WeChat group. Without the plaintiff’s consent, the defendant used AI software to generate an animated-style image depicting the plaintiff’s portrait (used as their WeChat profile picture) in revealing clothing and shared it in the group. Despite repeated objections from the plaintiff, the defendant continued to use the AI software to generate additional images of the plaintiff’s profile picture, portraying it with exposed clothing and bodily distortions, and sent these images directly to the plaintiff via private WeChat messages. The plaintiff argued that the infringing images, both shared in the group and sent privately, were recognizable as the plaintiff’s likeness and carried strong sexual connotations and derogatory implications, thereby reducing the plaintiff’s social standing and infringing upon their right of portrait, right of reputation, and general personality rights. The plaintiff sought a formal apology and compensation for moral and economic damages.
The court held that the infringing images posted in the WeChat group were generated without authorization using the plaintiff’s profile picture. These images exhibited a high degree of correspondence with the plaintiff’s natural appearance in terms of facial structure, posture, and style, enabling group members to identify the plaintiff based on the depicted physical characteristics and the context of the group chat. Thus, the defendant’s group sharing constituted infringement of the plaintiff’s right of portrait. The plaintiff’s profile picture, displayed on WeChat, served as an identifier of their virtual identity. The defendant’s AI-generated images, which altered the plaintiff’s appropriately dressed portrait into one with exposed breasts, triggered inappropriate discussions about the plaintiff within the group and objectively led to vulgar assessments of the plaintiff, infringing upon their right of portrait and right of reputation. Furthermore, the defendant used AI software to generate images depicting the plaintiff’s legs as wooden and even with three arms, which is obviously deviates from the basic human body structure, while also exposing the chest area. Sending such images directly to the plaintiff via private message was certain to cause psychological humiliation, violating the plaintiff’s personal dignity and constituting infringement of their general personality rights. The court ordered the defendant to issue a public apology to the plaintiff and compensate for moral damages. Neither party appealed the judgment, which has now taken effect.
Typical Case VII: A copyright ownership and infringement dispute between Mou Jia Technology Co., Ltd., Mou Yi Technology Co., Ltd. and Sun Moumou, Mou Network Technology Co., Ltd. – This case addresses the legal nature and originality assessment of virtual digital human images. A virtual digital human consists of external representation and technical core components, possessing digitalized appearance and anthropomorphic functions. At the external representation level, if the image reflects the production team’s unique aesthetic choices and judgments regarding lines, colors, and specific visual designs, satisfying the originality requirement for works, it may be recognized as a work of fine art protected under the Copyright Law.
Virtual Digital Humans A and B were jointly produced by four entities, including the first plaintiff, Mou Jia Technology Co., Ltd., and the second plaintiff, Mou Yi Technology Co., Ltd. the first plaintiff is the copyright owner, and the second plaintiff is a licensee. Virtual Digital Human A has over 4.4 million followers across platforms and was selected as one of the “Eight Major Cultural and Tourism Industry Highlights of 2022”. The plaintiffs asserted that the images of Virtual Digital Humans A and B constitute works of fine art, with Virtual Digital Human A’s image first published in the first episode of the short drama “Qian**” and Virtual Digital Human B’s image first published on the Weibo account “Zhi ****”. The first defendant, Sun Moumou, a former employee of one of the co-producing entities, unauthorizedly sold models of Virtual Digital Humans A and B on a model website operated by the second defendant, Mou Network Technology Co., Ltd., infringing upon the plaintiffs’ rights of reproduction and information network dissemination. The plaintiffs argued that the second defendant, as the platform operator, failed to fulfill its supervisory obligations and should bear joint liability with the defendant 1.
The court held that the full-body image of Virtual Digital Human A and the head image of Virtual Digital Human B were not directly derived from real persons but were created by the production team, demonstrating significant artistic effects. These images reflected the production team’s unique aesthetic choices and judgments in lines, colors, and specific visual designs, satisfying the originality requirement and constituting works of fine art. The infringing models published by the first defendant on the model website were identical or substantially similar to the protected works in terms of facial features, hairstyle, accessories, clothing design, and overall style, particularly in the combination of elements embodying the originality of the plaintiffs’ works, thus constituting substantial similarity and infringement of the plaintiffs’ right of information network dissemination. Considering factors such as the specific type of services provided by the second defendant, its level of intervention in the infringing content, whether it directly gained economic benefits, the reputation of the plaintiffs’ works, and the popularity of the infringing content, the court concluded that the second defendant, as an internet service provider, did not constitute joint infringement. Virtual digital humans carry multiple rights and interests, but this case focused solely on the rights pertaining to works of fine art. The amount of economic damages was determined based on the type of rights protected, market value, the infringer’s subjective fault, the nature and scale of the infringement, and the severity of the harm caused. The court ordered the first defendant to compensate the plaintiffs for economic losses. The first defendant appealed the first-instance judgment, but the appellate court dismissed the appeal and upheld the original ruling.
Typical Case VIII: An internet infringement dispute between He Mou and Mou AI Technology Company – This case establishes that a natural person’s personality rights extend to their virtual image, and unauthorized creation or use of such virtual images constitutes infringement of personality rights. Where a network service provider substantially participates in the generation and provision of infringing content through algorithmic design, it shall bear infringement liability as a content service provider.
The defendant, Mou AI Technology Company, is the developer and operator of a mobile accounting application that allows users to create “AI companions” by setting the companion’s name, avatar, and relationship with the user (e.g., romantic partner, siblings, parent-child). The plaintiff, He Mou, is a public figure with significant recognition and was extensively designated as a companion character by users of the application. When setting “He Mou” as a companion, users uploaded numerous portrait images of the plaintiff as avatars and defined relationship parameters. Through algorithmic deployment, the defendant categorized the companion “He Mou” based on relationship settings and recommended this character to other users. To enhance the anthropomorphic quality of AI characters, the defendant implemented a “training” algorithmic mechanism whereby users uploaded interactive corpora such as texts, portrait images, and dynamic emojis. Some users participated in reviewing this content, and the software filtered and categorized it to form character-specific corpora. Based on topic categories and character traits, the software could push portrait stickers and flirtatious dialogues during interactions between “He Mou” and users, creating an experience akin to real-life interaction with the plaintiff. The plaintiff alleged that the defendant’s actions infringed upon his right of name, right of portrait, and general personality rights, and thus filed a lawsuit.
The court held that the defendant’s actions constituted infringement of the plaintiff’s right of name, right of portrait, and general personality rights. Under the defendant’s software functionality and algorithmic design, users employed the plaintiff’s name and portrait to create virtual characters and develop interactive corpora, projecting the plaintiff’s overall identity—integrating name, portrait, and personality traits—onto AI characters, thereby forming a virtual image of the plaintiff. This constituted use of the plaintiff’s overall personality image, including their portrait and name. The defendant’s use of the plaintiff’s name and portrait without authorization infringed upon the plaintiff’s right of name and right of portrait. Furthermore, users could establish virtual relationships with AI characters, assign arbitrary forms of address, and “train” characters through corpora creation, resulting in a high degree of association between AI characters and the real natural person. This may evoke an emotional experience of genuine interaction with the plaintiff among users. Such use without the plaintiff's consent violated the plaintiff's personal dignity and freedom of personality, constituting infringement of general personality rights. Additionally, the services provided by the defendant’s software were fundamentally distinct from technical services. The defendant did not merely offer a platform for user content uploads but actively organized and encouraged users to develop infringing materials through rule-setting and algorithmic design, co-creating virtual images for use in user services. The defendant’s product design and algorithmic applications encouraged and facilitated the creation of the infringing virtual images and directly determined the core functionality of the software. Thus, the defendant was no longer a neutral technical service provider but liable as a network content service provider for infringement. The court ordered the defendant to issue a public apology to the plaintiff and compensate for moral and economic damages. The defendant appealed the first-instance judgment but later withdrew the appeal, rendering the judgment effective.
Typical Case I: A dispute involving infringement of authorship right and information network dissemination right between Li Moumou and Liu Moumou – This case establishes that content generated by natural persons using AI, if satisfying the definition of a work, shall be recognized as a work protected under the Copyright Law of the People’s Republic of China (“Copyright Law”). Ownership of copyright may be determined based on the original intellectual contributions of various participants, such as developers, users, and owners, to the AI-generated content.
On 24 February 2023, the plaintiff, Li Moumou, utilized the generative AI model Stable Diffusion to produce a photographic-style close-up portrait of a young woman under nightfall lighting conditions. Through selecting the model, inputting prompt and negative prompt words, and setting generation parameters, the plaintiff generated an image and published it on the Xiaohongshu platform on 26 February 2023. On 2 March 2023, the defendant Liu, Moumou, published an article on his Baijiahao account, using the image as an illustration. The defendant cropped out the plaintiff’s signature watermark and did not indicate the specific source of the image. The plaintiff thus filed a lawsuit, requesting a public apology and compensation for economic losses.
During court inspection, it was demonstrated that altering individual prompt words or parameters resulted in the generation of different images by the AI model. The court held that the image with a specific form of expression, in terms of its appearance, possesses indistinguishable from ordinary photographs or paintings and belongs to the field of art. The image was generated by the plaintiff using generative AI technology. From the conception to the final selection of the image, the plaintiff made substantial intellectual investments, thus rendering the image an “intellectual achievement”. The image exhibited identifiable differences from prior works, reflecting the plaintiff’s selections and arrangements. The process of adjustment and modification also demonstrated the plaintiff’s aesthetic choices and personal judgment. In the absence of evidence to the contrary, the image was independently completed by the plaintiff and embodied his personalized expression, thereby satisfying the requirement of “originality” for copyright protection. As a two-dimensional artistic work composed of lines and colors with aesthetic significance, the image qualifies as a work of fine art protected under the Copyright Law. Regarding copyright ownership, the Copyright Law stipulates that authors are limited to natural persons, legal persons, or unincorporated organizations. Therefore, the AI model itself cannot be considered an author under the Copyright Law. The plaintiff, as the natural person who configured the AI model according to specific needs and ultimately selected the image, is the author of the image entitled to copyright protection. The image was directly generated based on the plaintiff’s intellectual input and reflected their personalized expression. The defendant’s unauthorized use of the image as an illustration, making it accessible to the public at chosen times and locations, infringed upon the plaintiff’s information network dissemination right. Moreover, the removal of the plaintiff’s signature watermark violated the plaintiff’s authorship right. The defendant was thus liable for copyright infringement. The court ordered the defendant to issue a public apology and provide compensation for economic losses. Neither party appealed the judgment, which has now taken effect.
Typical Case II: A personal right infringement dispute between Yin Moumou and Mou Smart Technology Company et al. – This case establishes that a natural person’s right of voice extends to AI-generated voices and clarifies the criteria for determining whether voices processed by AI technology are protected under the right of voice.
The plaintiff, Yin Moumou, a voice actor, discovered that works generated using his voice recordings were widely disseminated on multiple well-known apps. Through voice screening and tracing, it was determined that the voices in these works originated from a text-to-speech product available on a platform operated by the first defendant, Mou Smart Technology Company. The plaintiff had previously been commissioned by the second defendant, Mou Cultural Media Company, to produce sound recordings, of which the second defendant is the copyright holder. The second defendant subsequently provided the plaintiff’s recordings to the third defendant, Mou Software Company. Using only one recording by the plaintiff as source material, the third defendant processed it through AI technology to develop the text-to-speech product in this case, which was then offered for sale on a cloud service platform operated by the fourth defendant, Mou Internet Technology Company. The first defendant entered into an online service agreement with the fifth defendant, Mou Technology Development Company, which placed an order with the third defendant that included the disputed text-to-speech product. The first defendant employed an application programming interface to directly access and generate the text-to-speech product on its platform without additional technical processing. The plaintiff claimed that the defendants’ actions seriously infringed his voice rights, and requested that the first and third defendants cease infringement and issue a public apology, and that all five defendants compensate the plaintiff for economic and moral damages.
The court held that a natural person’s voice, distinguished by its vocal print, timbre, and frequency, is unique, distinctive, and stable, capable of evoking associations or emotional responses related to that individual, thereby reflecting personal behavior and identity. Recognizability of a natural person’s voice means that upon repeated or prolonged exposure, the voice can identify a specific natural person. An AI-synthesized voice is considered recognizable if it enables the public or relevant audiences to associate the voice, based on its timbre, tone, and pronunciation style, with that natural person. In this case, as the third defendant developed the text-to-speech product exclusively using the plaintiff’s voice, and given that during in-court inspection the AI-generated voice was highly consistent with the plaintiff’s timbre, tone, and pronunciation style, it was capable of evoking associations with the plaintiff and identifying the plaintiff’s identity. While the second defendant held copyrights to the sound recordings, this did not include the right to authorize others to use the plaintiff’s voice for AI applications. The agreement between the second and third defendants, which authorized the AI use of the plaintiff’s voice without their informed consent, lacked legal basis. Thus, the defense of lawful authorization raised by the second and third defendants was rejected. The unauthorized AI use of the plaintiff’s voice by the second and third defendants constituted infringement of the plaintiff’s right of voice, resulting in harm for which they bear legal liability. The first, fourth, and fifth defendants were found subjectively free of fault and not liable for damages. Considering factors such as the nature of the infringement, market value of comparable products, and product usage metrics, damages were determined at the court’s discretion. The court ordered the first and third defendants to provide a written apology to the plaintiff, and the second and third defendants to compensate the plaintiff for economic losses. Neither party appealed the judgment, which has now taken effect.
Typical Case III: An internet infringement dispute between Li Moumou and Mou Cultural Media Co., Ltd. – This case establishes that unauthorized use of AI-synthesized celebrity voices for “live commerce” constitutes infringement. It clarifies that voices synthesized using AI deep synthesis technology fall within the protection scope of a natural person’s right of voice if they enable the public or relevant audiences to identify the specific natural person based on vocal characteristics such as timbre, tone, and pronunciation style. Additionally, in principal-agent relationships where merchants commission video promoters for sales, the merchant—as the commissioning party and actual beneficiary—bears a reasonable duty to review promotional content published by its associated video promoter. Failure to fulfill this duty results in joint liability with the video promoter.
The plaintiff, Li Moumou, possesses significant recognition and social influence in the fields of education and parenting. In 2024, the plaintiff discovered that the defendant, Mou Cultural Media Co., Ltd., utilized the plaintiff’s public speeches and instructional videos, paired with an AI-synthesized voice highly similar to the plaintiff’s, to promote multiple family education books sold through the defendant’s online store on a digital platform. The plaintiff alleged that the defendant’s unauthorized use of his portrait and AI-synthesized voice in promotional materials created a strong association between the plaintiff’s personal image and the defendant’s commercial promotions, misleading consumers into believing the plaintiff endorsed or recommended the books. By leveraging the plaintiff’s personal image, professional background, and social influence to attract attention and increase sales opportunities, the defendant infringed upon the plaintiff’s right of portrait and right of voice. As the book seller, the defendant maintained a principal-agent relationship with the video promoter (a live commerce streamer) to jointly conduct sales activities. The defendant possessed both the duty and capacity to review the promoter’s videos and was thus liable for the publication of the infringing content, including obligations to apologize and compensate for losses.
The court held that the video incorporated the plaintiff’s portrait and an AI-synthesized voice highly consistent with the plaintiff’s natural voice in timbre, tone, and pronunciation style. Given the plaintiff’s prominence in education and parenting, the video’s promotion of family education books made it likely for viewers to associate its content with the plaintiff. It was determined that relevant audiences could directly correlate the AI-synthesized voice with the plaintiff. Therefore, the voice fell within the protection scope of the plaintiff’s right of voice. The substantial use of the plaintiff’s portrait and synthesized voice without authorization constituted infringement of the plaintiff’s right of portrait and right of voice. The defendant and the video promoter had established a commissioned promotion relationship under platform rules and service agreements to jointly promote the defendant’s books and generate revenue. Based on platform rules and management authority, the defendant had the ability to review and manage the video content. Given the prominent use of the plaintiff’s portrait and synthesized voice, the defendant should have foreseen potential infringement risks and was obligated to reasonably verify whether the video had obtained the plaintiff’s authorization. Evidence indicated that the defendant failed to exercise due diligence in review. Thus, the defendant was held jointly liable with the video promoter for the publication of the infringing video. The court ordered the defendant to issue a formal apology to the plaintiff and compensate for economic losses and reasonable litigation costs. The plaintiff’s other claims were dismissed. Neither party appealed the judgment, which has now taken effect.
Typical Case IV: An internet infringement dispute between Liao Mou and Mou Science and Technology Culture Co., Ltd. – This case addresses the emerging business model of “AI face-swapping” and accurately distinguishes between portrait rights, personal information rights, and legally protected interests derived from creative labor in the application of generative synthetic AI. It establishes that unauthorized “AI face-swapping” of videos containing another person’s portrait constitutes infringement of personal information rights.
The plaintiff, Liao Mou, is a short-video blogger specializing in classical Chinese-style content with a substantial following across multiple platforms. Without authorization, the defendant, Mou Science and Technology Culture Co., Ltd., used a series of videos featuring the plaintiff to create face-swapping templates, which were uploaded to its software and offered to users for paid access, thereby generating profits. The plaintiff alleged that the defendant’s actions infringed upon his rights of portrait and personal information, demanding a written apology and compensation for economic and moral damages. The defendant contended that the videos published on its platform had lawful sources, that the facial features in the templates were not those of the plaintiff, and thus no portrait right infringement had occurred. Additionally, the defendant argued that the face-swapping technology was provided by a third party, and therefore it had not processed the plaintiff’s personal information.
The court found that the face-swapping template videos exhibited identical characteristics to the plaintiff’s original works in terms of makeup, hairstyle, clothing, movements, lighting, and scene transitions, but the facial features in the templates were entirely different and did not belong to the plaintiff. The defendant’s software utilized third-party services to implement the face-swapping function, and users could unlock all features by paying a membership fee. The court held that the determinative factor for portrait right infringement is recognizability. Recognizability emphasizes that the essence of a portrait lies in its ability to identify a specific natural person. While the scope of a portrait primarily centers on facial features, it may also include distinctive body parts, voices, or highly identifiable actions that correspond to a specific natural person. In this case, although the defendant used the plaintiff’s videos to create templates, it did not utilize the plaintiff’s portrait but instead replaced the facial features through technological means. Elements such as makeup, hairstyle, clothing, lighting, and scene transitions retained in the templates are not inseparable from a specific natural person and differ from inherent personal attributes. The public would identify the subject in the altered videos as a third party rather than the plaintiff. Moreover, the defendant’s provision of these templates to users did not involve defamation, distortion, or forgery of the plaintiff’s portrait. Thus, the defendant’s actions did not constitute infringement of the plaintiff’s right of portrait. However, the defendant collected videos containing the plaintiff’s facial information and replaced the plaintiff’s face with faces from other provided photos. This synthesis process required algorithmic integration of features from new static images with partial facial characteristics and expressions from the original videos. This procedure involved the collection, use, and analysis of the plaintiff’s personal information, constituting processing of personal information. The defendant’s processing of this information without the plaintiff’s consent infringed upon the plaintiff’s personal information rights. Regarding the unauthorized use of videos created by others, which may infringe upon rights derived from creative labor, such claims should be asserted by the relevant rights holders. The court ordered the defendant to issue a written apology to the plaintiff and compensate for moral damages and emotional distress. The plaintiff’s other claims were dismissed. Neither party appealed the judgment, which has now taken effect.
Typical Case V: A network service contract dispute between Tang Moumou and Mou Technology Co., Ltd. – This case establishes that online content platforms utilizing algorithmic tools to detect AI-generated content bear liability for breach of contract if they fail to fulfill reasonable and appropriate explanation obligations. While affirming the positive role of platforms in reviewing and managing AI-generated content through algorithmic tools to fulfill their information management responsibilities, the case clarifies that platforms must provide adequate explanations for automated decision-making results in scenarios involving real-time creative text.
The plaintiff, Tang Moumou, published a text of approximately 200 characters on a platform operated by the defendant, Mou Technology Co., Ltd, stating: “Work cannot truly earn you much money, but it can open new perspectives... If you are interested in learning to drive and plan to do so in the future, complete it during your leisure holiday... After starting work, there will be little uninterrupted time to obtain a driver’s license”. The defendant’s platform flagged the content as violating its policy for “containing AI-generated content without labeling”, hid the content, and imposed a one-day account suspension on the user. The plaintiff’s subsequent appeal was unsuccessful. The plaintiff claimed that the content was human-generated and that the defendant’s actions constituted a breach of contract, requesting the court to revoke the penalty, restore the content, and delete the violation record from the backend system.
The court held that users employing AI tools for content creation must label such content honestly and in accordance with the principle of good faith. The defendant had issued a community announcement requiring creators to actively label AI-generated content (AIGC) upon publication. The announcement stipulated that content without explicit labeling would be subject to circulation restrictions and retroactive labeling by the platform, and that unlabeled use of AIGC constituted a violation. As a registered user, the plaintiff was bound by this announcement, which formed part of the platform’s service agreement. The defendant was entitled to review and process user content to determine whether it was AI-generated. Generally, the plaintiff should provide preliminary evidence of human authorship, such as drafts, originals, source files, or source data. However, in this case, the plaintiff’s content was created in real time, making it objectively impossible to provide such evidence. The defendant’s conclusion that the content was AI-generated was based solely on algorithmic tool results. As the controller of both the algorithmic tool and the decision-making process, the defendant bore the responsibility to provide reasonable proof or explanation for its determination. Although the defendant submitted algorithmic filing information, it failed to demonstrate its relevance to the disputed content. The defendant did not adequately explain the basis or results of its algorithmic decision, nor did it justify its determination that the content was AI-generated. Consequently, the defendant was liable for breach of contract due to its lack of factual basis in penalizing the plaintiff’s account. The court further noted that the defendant’s manual review standards, which relied on subjective assessments of “distinct human emotional characteristics”, lacked scientific basis, persuasiveness, and credibility. The court ordered the defendant to restore the folded content and delete the relevant backend records. The defendant appealed the first-instance judgment but later withdrew the appeal, rendering the judgment effective.
Typical Case VI: An internet infringement dispute between Cheng Mou and Sun Mou – This case establishes that unauthorized use of AI software to mock or distort another person’s portrait constitutes infringement of personality rights. It emphasizes that users of generative AI technology must comply with laws, respect social morality and ethical principles, and refrain from endangering the legitimate rights and interests of others.
The plaintiff, Cheng Mou, and the defendant, Sun Mou, were members of the same photography discussion WeChat group. Without the plaintiff’s consent, the defendant used AI software to generate an animated-style image depicting the plaintiff’s portrait (used as their WeChat profile picture) in revealing clothing and shared it in the group. Despite repeated objections from the plaintiff, the defendant continued to use the AI software to generate additional images of the plaintiff’s profile picture, portraying it with exposed clothing and bodily distortions, and sent these images directly to the plaintiff via private WeChat messages. The plaintiff argued that the infringing images, both shared in the group and sent privately, were recognizable as the plaintiff’s likeness and carried strong sexual connotations and derogatory implications, thereby reducing the plaintiff’s social standing and infringing upon their right of portrait, right of reputation, and general personality rights. The plaintiff sought a formal apology and compensation for moral and economic damages.
The court held that the infringing images posted in the WeChat group were generated without authorization using the plaintiff’s profile picture. These images exhibited a high degree of correspondence with the plaintiff’s natural appearance in terms of facial structure, posture, and style, enabling group members to identify the plaintiff based on the depicted physical characteristics and the context of the group chat. Thus, the defendant’s group sharing constituted infringement of the plaintiff’s right of portrait. The plaintiff’s profile picture, displayed on WeChat, served as an identifier of their virtual identity. The defendant’s AI-generated images, which altered the plaintiff’s appropriately dressed portrait into one with exposed breasts, triggered inappropriate discussions about the plaintiff within the group and objectively led to vulgar assessments of the plaintiff, infringing upon their right of portrait and right of reputation. Furthermore, the defendant used AI software to generate images depicting the plaintiff’s legs as wooden and even with three arms, which is obviously deviates from the basic human body structure, while also exposing the chest area. Sending such images directly to the plaintiff via private message was certain to cause psychological humiliation, violating the plaintiff’s personal dignity and constituting infringement of their general personality rights. The court ordered the defendant to issue a public apology to the plaintiff and compensate for moral damages. Neither party appealed the judgment, which has now taken effect.
Typical Case VII: A copyright ownership and infringement dispute between Mou Jia Technology Co., Ltd., Mou Yi Technology Co., Ltd. and Sun Moumou, Mou Network Technology Co., Ltd. – This case addresses the legal nature and originality assessment of virtual digital human images. A virtual digital human consists of external representation and technical core components, possessing digitalized appearance and anthropomorphic functions. At the external representation level, if the image reflects the production team’s unique aesthetic choices and judgments regarding lines, colors, and specific visual designs, satisfying the originality requirement for works, it may be recognized as a work of fine art protected under the Copyright Law.
Virtual Digital Humans A and B were jointly produced by four entities, including the first plaintiff, Mou Jia Technology Co., Ltd., and the second plaintiff, Mou Yi Technology Co., Ltd. the first plaintiff is the copyright owner, and the second plaintiff is a licensee. Virtual Digital Human A has over 4.4 million followers across platforms and was selected as one of the “Eight Major Cultural and Tourism Industry Highlights of 2022”. The plaintiffs asserted that the images of Virtual Digital Humans A and B constitute works of fine art, with Virtual Digital Human A’s image first published in the first episode of the short drama “Qian**” and Virtual Digital Human B’s image first published on the Weibo account “Zhi ****”. The first defendant, Sun Moumou, a former employee of one of the co-producing entities, unauthorizedly sold models of Virtual Digital Humans A and B on a model website operated by the second defendant, Mou Network Technology Co., Ltd., infringing upon the plaintiffs’ rights of reproduction and information network dissemination. The plaintiffs argued that the second defendant, as the platform operator, failed to fulfill its supervisory obligations and should bear joint liability with the defendant 1.
The court held that the full-body image of Virtual Digital Human A and the head image of Virtual Digital Human B were not directly derived from real persons but were created by the production team, demonstrating significant artistic effects. These images reflected the production team’s unique aesthetic choices and judgments in lines, colors, and specific visual designs, satisfying the originality requirement and constituting works of fine art. The infringing models published by the first defendant on the model website were identical or substantially similar to the protected works in terms of facial features, hairstyle, accessories, clothing design, and overall style, particularly in the combination of elements embodying the originality of the plaintiffs’ works, thus constituting substantial similarity and infringement of the plaintiffs’ right of information network dissemination. Considering factors such as the specific type of services provided by the second defendant, its level of intervention in the infringing content, whether it directly gained economic benefits, the reputation of the plaintiffs’ works, and the popularity of the infringing content, the court concluded that the second defendant, as an internet service provider, did not constitute joint infringement. Virtual digital humans carry multiple rights and interests, but this case focused solely on the rights pertaining to works of fine art. The amount of economic damages was determined based on the type of rights protected, market value, the infringer’s subjective fault, the nature and scale of the infringement, and the severity of the harm caused. The court ordered the first defendant to compensate the plaintiffs for economic losses. The first defendant appealed the first-instance judgment, but the appellate court dismissed the appeal and upheld the original ruling.
Typical Case VIII: An internet infringement dispute between He Mou and Mou AI Technology Company – This case establishes that a natural person’s personality rights extend to their virtual image, and unauthorized creation or use of such virtual images constitutes infringement of personality rights. Where a network service provider substantially participates in the generation and provision of infringing content through algorithmic design, it shall bear infringement liability as a content service provider.
The defendant, Mou AI Technology Company, is the developer and operator of a mobile accounting application that allows users to create “AI companions” by setting the companion’s name, avatar, and relationship with the user (e.g., romantic partner, siblings, parent-child). The plaintiff, He Mou, is a public figure with significant recognition and was extensively designated as a companion character by users of the application. When setting “He Mou” as a companion, users uploaded numerous portrait images of the plaintiff as avatars and defined relationship parameters. Through algorithmic deployment, the defendant categorized the companion “He Mou” based on relationship settings and recommended this character to other users. To enhance the anthropomorphic quality of AI characters, the defendant implemented a “training” algorithmic mechanism whereby users uploaded interactive corpora such as texts, portrait images, and dynamic emojis. Some users participated in reviewing this content, and the software filtered and categorized it to form character-specific corpora. Based on topic categories and character traits, the software could push portrait stickers and flirtatious dialogues during interactions between “He Mou” and users, creating an experience akin to real-life interaction with the plaintiff. The plaintiff alleged that the defendant’s actions infringed upon his right of name, right of portrait, and general personality rights, and thus filed a lawsuit.
The court held that the defendant’s actions constituted infringement of the plaintiff’s right of name, right of portrait, and general personality rights. Under the defendant’s software functionality and algorithmic design, users employed the plaintiff’s name and portrait to create virtual characters and develop interactive corpora, projecting the plaintiff’s overall identity—integrating name, portrait, and personality traits—onto AI characters, thereby forming a virtual image of the plaintiff. This constituted use of the plaintiff’s overall personality image, including their portrait and name. The defendant’s use of the plaintiff’s name and portrait without authorization infringed upon the plaintiff’s right of name and right of portrait. Furthermore, users could establish virtual relationships with AI characters, assign arbitrary forms of address, and “train” characters through corpora creation, resulting in a high degree of association between AI characters and the real natural person. This may evoke an emotional experience of genuine interaction with the plaintiff among users. Such use without the plaintiff's consent violated the plaintiff's personal dignity and freedom of personality, constituting infringement of general personality rights. Additionally, the services provided by the defendant’s software were fundamentally distinct from technical services. The defendant did not merely offer a platform for user content uploads but actively organized and encouraged users to develop infringing materials through rule-setting and algorithmic design, co-creating virtual images for use in user services. The defendant’s product design and algorithmic applications encouraged and facilitated the creation of the infringing virtual images and directly determined the core functionality of the software. Thus, the defendant was no longer a neutral technical service provider but liable as a network content service provider for infringement. The court ordered the defendant to issue a public apology to the plaintiff and compensate for moral and economic damages. The defendant appealed the first-instance judgment but later withdrew the appeal, rendering the judgment effective.