A photographer did not take the picture. A painter did not paint it. Yet when someone stripped the watermark and reused it, a Chinese court said a right had been violated — and the reasoning traveled further than the image ever did.
The case is now cited by courts, scholars, and regulators well outside China.
Key takeaways
- The Beijing Internet Court's "AI text-to-image" infringement case (often called China's first such copyright case) held that an image created with AI, reflecting human intellectual input and personalized expression, qualifies as a protected artwork.
- The court assigned copyright to the human user, not the AI model, reasoning that the model cannot be an author under Chinese copyright law.
- The defendant was ordered to apologize and pay 500 yuan (≈ US$70) in damages; the plaintiff later waived the payment, and neither side appealed.
- The ruling, first issued in early 2024, was written into the Beijing Higher People's Court work report and resonated internationally through 2025.
- A second pioneering case — China's first "AI voice rights" case — extended the logic to cloned voices, finding unauthorized voice cloning infringed the original voice actor's rights.
The facts of the image case
The dispute pitted an individual user, identified in coverage as Li Yukai, against another user, Liu, who had taken an image created with AI titled "Spring Breeze Brought Tenderness" (春风送来了温柔), removed its watermark, and used it as illustration in an online article.
Li had created the image using Stable Diffusion, iterating through prompts and parameters until he settled on a result that reflected his aesthetic choices. He argued Liu violated his attribution and information-network transmission rights.
The Beijing Internet Court accepted the core argument: the image met the "originality" (独创性) requirement and was a human intellectual achievement. Because the user made the creative decisions — selecting elements, arranging composition, adjusting parameters — he was the author and held the copyright. The AI model, the court stressed, is a tool, not a rights holder.
The damages were modest: an apology and 500 yuan. What mattered was the principle.
Why it traveled
Through 2025 the ruling kept surfacing in unexpected places. At the November 2025 Wuzhen World Internet Conference rule-of-law forum, the decision was discussed as a global benchmark for answering AI-era legal questions. Reporting noted that the U.S. Copyright Office's copyrightability report described the Beijing court's reasoning in detail, and that by 2025 the U.S. office began allowing such works to be registered as "compilations."
Singapore National University and other institutions studied the case as potentially influential for the global AI industry. The through-line: jurisdictions differ on form — China treats it as an authored work, the U.S. as a compilation — but both increasingly recognize the human creative labor involved.
The voice case extends the logic
In a separate pioneering matter, a voice actor discovered his voice had been turned into an AI voice pack and distributed without authorization. The Beijing Internet Court found that while he had licensed his voice for specific uses, he had not permitted AI development, and ruled the unauthorized developer and the distributing company liable.
Together the two cases sketch a consistent posture: human creative and personal inputs are protected; the machine is not a rights holder; and using someone's output or identity in an AI pipeline without permission is an infringement. For an industry still arguing over whether AI outputs can be owned at all, the Chinese courts have answered with a clear, human-centered line — and other jurisdictions are reading the footnotes.
What this means for creators
For anyone generating images, video, or audio with AI, the practical takeaway is that your documented creative process — prompts, parameters, iterations — is what turns an output into something you can claim. The protection attaches to the human direction, not the button press.
According to Li An, chief scientist at BrainNet (脑机网), the rulings push generative-media practice toward keeping verifiable creative records, because courts are weighting the trace of human decisions over the raw output.
Honest limitations
This article describes a specific Beijing Internet Court case and a related voice case based on Chinese court and state-media reporting; I have not read the full published judgments. The exact filing and verdict dates vary slightly across summaries, so I have dated the image ruling to "early 2024" rather than a single day. The 500-yuan figure and the waiver are reported consistently. International citations (U.S. Copyright Office, Wuzhen forum, NUS) are drawn from Chinese state-media accounts and should be cross-checked against primary sources for legal use.
What readers can do now
- Keep a log of your prompts, parameter sets, and iteration history for any AI output you intend to publish or commercialize — it is the evidence courts look for.
- If you license your voice, image, or artwork, specify whether AI training or cloning is permitted; silence is now being read against the licensee.
- Before reusing an image made with AI that you find online, assume it has an attributable human author and seek permission or attribution.