Appendix B: Li vs Liu (2023)¶

B.1. Points analyzed by the court¶

The Beijing Internet Court examined the following key points to reach its decision:

Evaluation of the copyrightability of the work: the image falls within the field of visual art. It is the result of Li's intellectual effort, who used AI as a creative tool. It possesses originality, reflecting Li's creative choices in the selection and arrangement of visual elements. The use of technological tools does not preclude copyright protection, provided there is significant human creative contribution.

Determination of authorship: Li had an active role in the creative process, providing specific inputs to the AI system. Li exercised significant control over the AI's output through the selection of prompts, parameters, and iterations. Li made creative decisions in the selection and refinement of the final image. The AI was used as a tool, analogous to a camera or editing software, and not as an autonomous creator.

Classification of the work: the image satisfies the criteria defined in the Regulations for the Implementation of the Copyright Law for fine art works. It is a two-dimensional visual work composed of lines and colors. It possesses recognizable aesthetic value.

Evaluation of copyright infringement: Liu used the image without authorization, violating the right of dissemination on computer networks. The removal of the watermark was considered a violation of the moral right of authorship of the work. Liu could not invoke ignorance of the source of the image as justification.

Determination of sanctions: Liu must publish an apology to repair the damage to Li's reputation as an author. Compensation was established based on the difficulty of determining actual damage or illicit gain. The court considered the circumstances of the violation and the impact on the rights holder.

Considerations on disclosure of AI use: authors should disclose the use of AI, but it is not a binding legal requirement. Disclosure promotes transparency and allows the public to understand the creative process. The court recognized the importance of informing the public about the new technologies used in artistic creation. A legal obligation was not imposed, recognizing that AI is a creative tool like others.

B.2. Chronology of actions taken by Li¶

The court analyzed in detail Li's creative process, identifying the following actions:

Software selection: Li chose to use Stable Diffusion, an open source AI software.

Model choice: Li downloaded and used two specific models: "AsiaFacemix-pruned-fix.safetensors" and "lora-hanfugirl-v1-5.safetensors."

Setting the textual prompt: Li composed a detailed prompt that included:

Artistic type: "ultra photorealistic," "color photo"
Subject: "japan idol" with detailed descriptions of skin, eyes, braids
Setting: "in locations," "golden hour," "dynamic lighting"
Composition: "cool pose," "viewing at camera"
Style: "film texture," "film simulation"

Setting the negative prompt: Li included elements to exclude, such as "3d, render, cgi, painting, drawing, cartoon, anime, comic."

Adjustment of technical parameters:

Number of "iteration steps": set to 33
Image height: set to 768 pixels
"CFG scale": set to 9
"Random seed": initially set to 2692150200

Generation and selection of images: Li generated multiple versions of the image. They selected the most promising versions for further iterations.

Parameter modifications for subsequent iterations: modification of the weight of "lora-hanfugirl-v1-5.safetensors" to 0.75. Change of "random seed" to 2692150199.

Prompt refinement: addition of terms such as "shy, elegant, cute, lust, cool pose, teen, viewing at camera, masterpiece, best quality."

Final image selection: Li chose the final image from among the hundreds generated.

Post-processing (mentioned but not detailed in the ruling): possible use of additional software for final touch-ups.

Publication: Li published the final image on the social platform Little Red Book.

This detailed chronology of Li's actions was fundamental to the court's decision to recognize their creative contribution and, consequently, their authorship of the work generated with the assistance of AI.