This case marks a significant milestone as the first national ruling aimed at safeguarding the copyright of AI-generated “text-to-image” works.
In its ruling, the Chinese court maintained an “anthropocentric” approach to defining copyright ownership, viewing the AI model as a supportive tool rather than the primary creator. It underscores the legislative intent of “encouraging creation” while reaffirming the primary role of “humans” in the creative process, positioning AI as a supportive tool. An analysis of court decisions before and after this landmark case indicates a notable shift in China’s judicial approach to the copyrightability of AI-generated content (AIGC).
In this case and in subsequent rulings, most courts are now inclined to perform detailed, case-by-case evaluations to assess the elements of “intellectual input” and “originality.”
Summary of Judicial Viewpoints
In relation to the classification of AIGC images as “works,” the court emphasizes the importance of assessing “intellectual achievements” and “originality.” When generating images with AI tools, if the user’s specific guidance and adjustments—such as selecting models, entering prompt words, and configuring parameters—demonstrate their intellectual contribution, the resulting images will be acknowledged as the user’s “intellectual achievements.” Additionally, if the unique generation process and outcomes can substantiate a “personalized expression,” the images will be deemed to fulfill the “originality” criteria established by copyright law.
In addressing the question of who qualifies as the “author” of images generated by artificial intelligence (AIGC), it’s important to recognize the various parties involved in the image creation process. These include the AI model itself, the designer of the AI model, and the user who operates the model. According to the Copyright Law, authorship is restricted to individuals or entities that can be classified as civil subjects, which means that the AI model cannot be considered an author. Additionally, while the AI designer contributes to the development of the tool, their intellectual input is not directly tied to the creation of the image itself, and therefore they cannot be recognized as the author either. Courts typically evaluate factors like the user’s intellectual contribution and unique expression, along with relevant documents like licensing agreements, to conclude that the user is the author.
Case Overview
Mr. Li, the plaintiff, created the image “Spring breeze brings gentle tender” using the open-source software Stable Diffusion by entering specific prompt words and shared it on the Xiaohongshu platform. Subsequently, Mr. Liu, the defendant, wrote an article titled “Love in March, Amidst the Peach Blossoms,” which included the disputed images as illustrations without permission and removed the watermark that credited Mr. Li on the Xiaohongshu platform. Mr. Li contends that Mr. Liu’s actions have caused users to mistakenly believe that Mr. Liu is the original creator of the work, significantly violating Mr. Li’s rights to authorship and the dissemination of information online. As a result, Mr. Li has initiated legal proceedings against Mr. Liu.
The plaintiff obtained the Stable Diffusion model and input numerous prompt words into both the positive and negative prompt fields. They also adjusted the number of iterations, image height, prompt guidance coefficient, and random seed to create Image 1. With the same parameters, altering the weights of one of the models produced Figure 2. Again, keeping the parameters the same, changing the random seed resulted in Image 3. Provided that the aforementioned parameters remain consistent, the addition of further content to the positive prompt words resulted in the creation of image 4 (the disputed image).
Court Rulings
When discussing “intellectual achievements,” we refer to the results of intellectual endeavors.
Consequently, a work should embody the intellectual contributions of an individual. From the initial conception of the image to its final selection, the plaintiff engaged in various intellectual investments throughout this process. These included designing character presentations, selecting prompt words, organizing the sequence of those words, setting relevant parameters, and determining which image aligned with their expectations.As such, the resulting image reflects the plaintiff’s intellectual investment, thereby qualifying as an “intellectual achievement.”
Intellectual achievements that demonstrate “originality” are essential for defining works.
In general, “originality” implies that the work is independently created by the author and showcases their unique expression. When it comes to determining whether the use of artificial intelligence to create images reflects an author’s personalized expression, this assessment must be made on an individual basis rather than applied broadly. Typically, when individuals utilize models like Stable Diffusion to generate images, the uniqueness of their requests and the clarity and specificity of their descriptions regarding the elements and layout of the image can significantly enhance the reflection of their personal expression.
How do intellectual achievements and originality manifest in this particular case?
From the perspective of the disputed image, it exhibits distinct differences from previous works. Although the plaintiff did not manually draw specific lines or provide detailed instructions to the Stable Diffusion model regarding the creation of lines and colors, it can be said that the model effectively “drew” the lines and colors that make up the image. This approach differs significantly from traditional methods that involve using brushes or drawing software.
The plaintiff played a crucial role in designing the image’s elements, such as the characters and their presentation, by utilizing prompt words. Additionally, they established the layout and composition through specific parameters, showcasing their creative choices and arrangements. After generating the initial image by inputting prompt words and setting relevant parameters, the plaintiff further refined the image by adding new prompt words and adjusting parameters, ultimately leading to the final version. This iterative process highlights the plaintiff’s aesthetic preferences and individual judgments.
During the trial, the plaintiff demonstrated the model’s versatility by generating various images through minor adjustments to prompt words or parameters. This indicates that different users can input unique prompt words and set new parameters to create diverse content. Consequently, the image in question cannot be classified as a mere “mechanical intellectual achievement.” In the absence of any contrary evidence, it is reasonable to conclude that the image was independently created by the plaintiff and reflects their personalized expression. Therefore, it meets the requirement of “originality.”
In addressing the question of authorship for images generated by artificial intelligence, it is important to recognize that the creative process fundamentally involves human input.
When individuals utilize AI models to create images, the intellectual contributions come from the humans operating these tools, rather than the AI itself. The primary goal of the copyright system is to foster creativity, and it is essential to apply this system effectively to encourage broader use of innovative tools in the creative process.
Given the current technological landscape, images produced by AI that incorporate original human intellectual input should be acknowledged as works deserving copyright protection. In this case, the plaintiff actively configured the parameters of the AI model to meet specific needs and ultimately selected the resulting image. This image was a direct outcome of the plaintiff’s intellectual contributions and personal expression, establishing the plaintiff as the author and copyright holder.
While the court has recognized the plaintiff as the author with copyright rights in this instance, it is also important to uphold transparency and the public’s right to know. Therefore, the plaintiff is encouraged to clearly indicate the use of AI technology or models in the creation process. In this case, the plaintiff appropriately labeled the work as “AI illustration,” which effectively informs the public that the content was generated using artificial intelligence. The court supports this approach.
Takeaways for Providers of Generative Artificial Intelligence Services
The creators of the Stable Diffusion model have explicitly stated in their licensing agreement that they do not assert any rights over the outputs produced by users of the model. Instead, users are accountable for the outputs they generate and how they choose to utilize them. This clarification helps establish the rights and responsibilities associated with generated content, providing a crucial reference point for courts when determining copyright ownership.
To ensure compliance and minimize potential disputes, AI service providers should align their practices with relevant regulations, including the “Regulations on the Management of Algorithm Recommendation for Internet Information Services,” the “Interim Measures for the Management of Generative Artificial Intelligence Services,” and the “Regulations on the Management of Deep Synthesis for Internet Information Services.”
It is essential for providers to clearly outline the rights and obligations of users in their agreements. Additionally, they should strive to label generated content prominently in accordance with the upcoming “Measures for the Identification of AI-Generated Synthetic Content,” which will take effect in September 2025, along with the associated mandatory national standards.