Beijing court grants copyright to an AI-generated image, splitting with the US
The Beijing Internet Court credited the prompter, not the AI or its developer, with authorship — the opposite conclusion the US Copyright Office had reached on Stable Diffusion images months earlier.
- Courts & copyright
- Notable
The Beijing Internet Court ruled, in Li v. Liu, that an image the plaintiff had generated using Stable Diffusion qualified for copyright protection — the first Chinese court decision to extend copyright to an AI-generated image. The defendant, a blogger who had used the plaintiff’s image without permission, argued that a machine-generated picture could not be a copyrightable “work” at all; the court disagreed.
Applying China’s existing four-element test for what counts as a protectable work, the court found the image original and the product of “intellectual achievement” attributable to a human. It weighed the plaintiff’s process — selecting and refining more than 150 prompts, setting technical parameters and iterating across multiple outputs before settling on the final image — as evidence of substantial creative input, distinct from simply clicking a button and accepting whatever a model produced. Authorship, the court held, belonged to the person directing that process, not to the AI model or the company that built it.
The ruling put Chinese and American law on opposite sides of the same underlying question. Months earlier, the US Copyright Office had stripped protection from individual Midjourney images in the graphic novel Zarya of the Dawn, reasoning that a prompter could not sufficiently control a diffusion model’s output to count as its author — and a Washington court would go further the following year, holding that a work with no human author at all could not be copyrighted regardless of the prompting involved. The Beijing court treated essentially the same prompt-and-iterate workflow as sufficient human authorship; the US position treated it as insufficient.
The divergence mattered beyond the individual case: it meant the same AI-generated image could be protectable in one major jurisdiction and unprotectable in another, leaving companies operating across both without a single settled answer to who, if anyone, owns an AI-assisted image.