Chinese court declares that AI-generated image has copyright

The copyright law world has been abuzz last week with news that a court in the People’s Republic of China decided that a work generated with Stable Diffusion had copyright, and therefore the author could sue for copyright infringement.

The case is Li v Liu (page with full verdict here), in which a person named Li used an AI image generator to produce an image of a young woman. A blogger named Liu took the image and used it in a blog post, Mr Li sued for copyright infringement, and won CN¥500, with costs of CN¥50 (just over £61 in total, or $76).

The dispute began when Mr Li generated pictures using an AI service, Stable Diffusion, and posted one titled “Spring Breeze Brings Tenderness” on the Chinese platform Little Red Book. Ms Liu, without Li’s permission, used this picture in her blog post after removing Mr Li’s user ID and the watermark. Mr Li subsequently sued for copyright infringement, and won.

The Internet Court had to handle three main questions. Firstly, the copyrightability of the AI-generated pictures, the court ruled that the picture met the criteria for copyright protection under Chinese law, including originality and being a work of art. The court emphasised that the picture was not merely a product of AI but reflected Mr Li’s intellectual contributions and aesthetic choices. Secondly, the authorship of the image, and here the court determined that the AI service or its developers were not the authors. Instead, Mr Li is the author because he provided significant intellectual input and personalized expression in creating the picture. Thirdly, as there was copyright the defendant was liable for copyright infringement for using the picture without authorisation and altering it by removing identifying marks.

When discussing the issue of whether an AI-generated work can be subject to copyright, the court decided to bypass the actual technological aspect of the generation, and concentrated on the level of human involvement. The court says:

“From the time when the plaintiff conceived the picture involved in the case to the final selection of the picture involved, the plaintiff has made a certain amount of intellectual investment, such as designing the presentation of characters, selecting prompt words, and arranging The order of prompt words, setting relevant parameters, selecting which picture meets expectations, etc. The picture involved reflects the plaintiff’s intellectual investment, so the picture involved meets the requirements of “intellectual achievement.””

And regarding originality, the court continues:

“The plaintiff designed the characters and their presentation and other picture elements through prompt words, and set the picture layout and composition through parameters, which reflected the plaintiff’s choices and arrangements. On the other hand, after the plaintiff obtained the image he continued to add prompt words, modify parameters, continuously adjusting and modifying, and he finally obtained the picture involved. This adjustment and modification process also reflects the plaintiff’s aesthetic choice and personality judgment… The pictures involved in the case are not “mechanical intellectual achievements.” In the absence of contrary evidence, it can be determined that the pictures involved in the case were independently completed by the plaintiff and reflect the plaintiff’s personalized expression. In summary, The pictures involved in the case meet the requirements of “originality”.” (translated with DeepL and edited for clarity).

An important part of this analysis was the evidence presented, both of selection of prompts, as well as the selection of outputs. The prompt used was:

“ultra-photorealistic: 1.3), extremely high quality highdetail RAW color photo, in locations, Japan idol, highly detailed symmetrical attractive face, angular symmetrical face, perfect skin, skin pores, dreamy black eyes, reddish-brown plaits hairs, uniform, long legs, thighhighs, soft focus, (film grain, vivid colors, Film emulation, kodak gold portra 100, 35mm, canon50 f1,2), Lens Flare, Golden Hour, HD, Cinematic, Beautiful Dynamic Lighting”

And the author also included 120 negative prompts, as well as later modifying parameters and using seeds to further narrow down the output. All of these steps were used by the court as evidence that a sufficient amount of intellectual achievement had been reached. The originality standard in China being that a work needs to show an intellectual achievement, this is translated as an author’s aesthetic choice and personality, transmitting the human author’s personalised expression.

Evolution of the outputs, first on the left, final on the right.

It’s also important to stress that this is not the first time that a Chinese court has found copyright authorship in a work, in 2020 a court also found that an article that had been written by an AI had copyright (more of that here and in this article by Yours Truly).

What next? Those who do not want to see the rise of the Copyright Androids do not need to worry about global implications of the ruling. Civil law countries such as China do not have a system of precedent like Common law systems, so other courts in the PRC can decide differently; the Beijing Internet Court is also a lower court, so the decisions can be overturned. However, we’ve now had two decisions in the same jurisdiction willing to give AI works copyright, so we could start seeing a trend. Courts can cite previous decisions and look at jurisprudential trends, so it will be interesting to see if this trend spreads, or if it’s a one-off. ETA: There is strong indication that this is actually not a one-off. The Supreme People’s Court, which is the top court in China, often selects model cases that can be used by the lower courts in their decisions in important and upcoming subjects. In 2021 the court chose the Tencent decision (which grants an AI copyright) as one of its 50 model cases. So this could become a trend.

Implications outside of the PRC

This decision is likely to become famous, and I can already hear the clacking of a thousand keyboards as students include this in their essays. (note to students: this could be a great dissertation topic).

While this decision will not set a global precedent, it raises important points for the wider copyright AI debates worldwide. The first point is its clear reversal of the trend set by the US Copyright Office regarding AI-generated works’ copyrightability. Over the past year, a meme has spread that AI works aren’t subject to copyright protection, based on the USCO’s decisions in the cases of Thaler and Zarya of the Dawn. However, the USCO is an outlier in international copyright law. No other country has an obligatory registration system (some voluntary registries exist); copyright exists on original works regardless of registration, so no other country has the same level of gatekeeping. It will be up to national courts to interpret this issue on a case-by-case basis. Moreover, the originality standard in the US is famously strict, making it even more of a potential outlier. I won’t delve into the issue that most people tend to consider any US decision as final around the globe; after all, the world is quite large.

Another aspect that I think will be relevant going forward is that this ruling contains arguments that could easily be adapted to the UK and Europe with a few tweaks in language. The UK, under s9(3) of the CDPA, allows for authorship of works generated by a computer without human intervention, a provision also present in countries such as India, New Zealand, Ireland, and South Africa. Assuming that the work also needs to meet the local originality requirement – that it must be an intellectual creation – this is quite similar to the Chinese ‘intellectual achievement’ standard.

What is an intellectual creation you ask? Interestingly, we recently got a very good analysis of what constitutes an intellectual creation in software development where the author seemingly has done little work. In the case of THJ Systems Ltd v Sheridan [2023] EWCA Civ 1354, Lord Justice Arnold discussed whether a software developer owned copyright in a graphic user interface. He explained that the originality test in the UK is that a work has copyright if it is the “author’s own intellectual creation”, and that while this test sets the bar higher to the old “skill and labour” test, “even a simple portrait photograph” can satisfy it as long as it meets the requirements (at 23). Discussing what is the level of creativity needed for protection in a GUI and some charts, Arnold LJ comments that it “is plain that the degree of visual creativity which went into the R & P Charts was low. But that does not mean that there was no creativity at all. The consequence of the low degree of creativity is that the scope of protection conferred by copyright in the R & P Charts is correspondingly narrow, so that only a close copy would infringe.”

Now take all of this, and look at the level of human action that can go into the creation of an AI-generated work, and I don’t think that a blanket prohibition works. Given enough author’s own intellectual creation, an AI work could very well be subject to protection. Selection of inputs and outputs could very well meet that threshold.

Concluding

The Chinese ruling is evidently a stand-alone decision, but I’d argue that the language used is compatible with the intellectual creation originality test, and it could prove to be the first of many such cases. I don’t think that every single AI-generated work has copyright, but some may very well do if they cross the threshold of originality. An author wrote back in 2017:

“What remains clear is that as UK courts have to meet the intellectual creation standard, then it is vital to look at each work on a case by case basis. Some works may have enough input from a human author to meet that requirement, and some may not.”

I wonder what happened to that guy.