The Beijing Internet Court recently decided an interesting copyright infringement lawsuit concerning AI generated images (GW law translated version here). The court held that the defendant’s use (which was of a non-commercial nature) of an image generated using AI (created by the plaintiff using Stable Diffusion) violated the plaintiff’s copyright. The defendant was asked to apologise and pay the plaintiff CNY 500.
The plaintiff created an image called ‘Tenderness Sent by the Spring Breeze’ using Stable Diffusion (below). The final image was created after providing several prompts. See here to understand how the final image was created.
The plaintiff then discovered that the defendant had used this image alongside a poem – ‘Love in March, amongst Peach Blossoms’. The image was published without obtaining the plaintiff’s permission and without the plaintiff’s watermark.
Original work by a human
The court held that the image qualifies as an original work of human intellectual achievement under applicable Chinese copyright law. Through various prompts, the plaintiff designed the character, the composition and layout of the image. These prompts reflected the plaintiff’s personal judgement, choice and arrangement. The work was therefore held to be the outcome of the intellectual effort of the plaintiff – a natural person. The court also states that this is a case-by-case assessment.
Further, the work was original since the creation of the AI generated image was not held to be a mechanical exercise involving the application of given sequences, formulas or structures. Rather, the court held that generation of the image by the plaintiff involved independent thought and personal judgement (through the prompts). This was also evidenced by the fact that different images can be created with different prompts.
IPKat questions the mechanical exercise reasoning here. It is also worth noting that some negative prompts were copied and pasted by the plaintiff from anonymous user’s shared content in an online forum.
Thoughts on commissioned works
The court notes that in the process of generating the image, the plaintiff did not ‘draw the specific lines or fully direct the Stable Diffusion model on how to draw these lines and colours’. Instead, the lines and colours were primarily ‘drawn’ by the Stable Diffusion model. In this regard, the court characterises the AI model as similar to that of a human being who has gained skills through training and learning and is capable of drawing on behalf of humans. Similar to a commissioned artist.
Having said this, the court says that there are major differences between commissioned works and AI generated works. The differences being one of free will and legal personality. While a commissioned artist has their own will to exercise judgement and choice, an AI model does not. And an AI model lacks legal personality.
On the lack of free will and judgement, the US Copyright Office appears to have a different view. In its decision on whether prompts qualify as sufficient human input, the USCO observed that prompts are not ‘direct instructions’ which AI models understand like humans do. Also, outputs are a result of the AI model’s interpretation of the prompts and its training dataset.
In a sense, AI models may be exercising some ‘free will’ / judgement shaped by prompts and training datasets. At the same time, the free will of a commissioned artist is not entirely free since it is shaped by the instructions provided by the person commissioning the work.
That said, both jurisdictions still require human authorship.
Other developments
The USCO has issued guidelines on ‘de minimis’ use of AI (in such instances disclosures or disclaimers are not required) (here, here, here). The guidelines also include a list of illustrations to help understand what kind of uses would qualify as de minimis. As reported, the Review Board of the USCO has refused to register a work created using AI four times as of December 2023.
Interestingly, the most recent refusal was for a work Suryast naming an AI tool Raghav as the creator. We covered Sahni’s attempt at the USCO here. Sahni also attempted to register an image created by the same tool before the Indian Copyright Office (here and here). Mr. Sahni has also successfully secured copyright registration for ‘Suryast’ in Canada, where an author of a protected work must necessarily be human. Speaking about this, Mr. Sahni said that his creativity and skills were inseparable from the contribution of the AI tool, and therefore, they were co-authors, or joint authors under the Canadian law.
In China, this is not the first case involving copyright and works generated using AI. There were two other cases: Feilin v. Baidu (Beijing Internet Court, 2019) and Shenzhen Tencent v. Shanghai Yingxun (Shenzhen Nanshan District People’s Court, 2019) outcomes of at least one of which are different from the Li v. Liu case.