Many artists allege that popular generative AI services violated copyright law by training on a dataset that included their works and, in some cases, that users of these services can directly reproduce copies of the work. Last year, Judge William Orrick allowed a direct copyright infringement complaint against Stability, operator of the popular Stable Diffusion AI image generator. But he dismissed a variety of other claims and asked the artists’ attorneys to amend them with more detail.
In this more recent ruling, the revised arguments convinced the judge to approve an additional claim of induced copyright infringement against Stability. He allowed a copyright claim against DeviantArt, which used a model based on Stable Diffusion, as well as against Runway AI, the initial startup behind Stable Diffusion. And he allowed copyright and trademark infringement claims against Midjourney.
The latter claims include allegations that Midjourney misled users with a “Midjourney Style List,” which included 4,700 artists whose names could be used to generate works in their style. The artists argue this list — created without their knowledge or approval — implies a false endorsement, and the judge found the accusation substantive enough to merit further argument.
Judge Orrick remained unconvinced by some of the arguments he had previously sent back for more detail. He threw out claims that the generators violated the Digital Millennium Copyright Act by removing or altering copyright management information. He also dismissed a claim that DeviantArt had breached its terms of service by allowing users’ work to be scraped for AI training datasets. And, obviously, the claims he did allow will still need to be argued in court.
Kelly McKernan, one of the artists behind the suit, described the ruling as “very exciting” and “a HUGE win” on X. McKernan noted that passing this preliminary stage lets them request information from companies in discovery — potentially revealing details about software tools that often remain black boxes. “Now we get to find out alllll the things those companies don’t want us to know,” McKernan wrote. (If the companies are ordered to produce information, it wouldn’t necessarily be released to the public.)
But the case’s outcome is difficult to predict. Numerous suits have been filed against AI companies, alleging that tools like Stable Diffusion and ChatGPT easily reproduce copyrighted works and are illegally trained on huge volumes of them. The companies have countered that these reproductions are rare and difficult to produce, and they argue that training should be considered legal fair use. Some early suits have been thrown out, including a GitHub Copilot case whose dismissal is mentioned in yesterday’s ruling. Others, like a New York Times Company suit against OpenAI, remain ongoing.
At the same time, OpenAI, Google, and other tech giants have struck multimillion-dollar deals with publishers (including Verge parent Vox Media) and photo providers for ongoing data access. Small companies like Stability and Midjourney have less money to buy access to data, and individual artists have less leverage to demand payments — so for both sides of this dispute, the legal stakes are particularly high.
