Many artists have argued that popular generative AI services violate copyright law by training them on datasets that include their works, and that in some cases users of these services can directly reproduce copies of their works. Last year, Judge William Orrick allowed a direct copyright infringement claim against Stability, the operator of the popular Stable Diffusion AI image generator. But he dismissed various other claims, asking the artists' lawyers to amend their case with more detail.
In this recent decision, with amended arguments, the judge granted additional inducement of copyright infringement claims against Stability. The judge granted copyright infringement claims against DeviantArt, which used a model based on Stable Diffusion, and Runway AI, the original startup behind Stable Diffusion. The judge also granted copyright and trademark infringement claims against Midjourney.
The latter claim includes allegations that MidJourney misled users with its “MidJourney Style List,” which included 4,700 artists whose names could be used to produce work in their style. The artists argue that the list, which was created without their knowledge or approval, represented a false endorsement, and the judge found the accusations to be substantial enough to merit further argument.
Judge Orrick was unconvinced by some of the arguments he previously remanded for more details. He rejected the argument that Generator violated the Digital Millennium Copyright Act by removing or altering copyright management information. He also rejected the argument that DeviantArt violated its terms of service by allowing users' works to be scraped for AI training datasets. And, of course, the arguments he did accept will still need to be argued in court.
Kelly McKernan, one of the artists who filed the lawsuit, called the ruling “very exciting” and a “huge win” for X. McKernan noted that getting through this preliminary stage means discovery can demand information from the companies, potentially revealing details about their software tools, which often remain black boxes. “Now we know everything the companies don't want us to know,” McKernan wrote. (Even if a company is ordered to turn over information, it doesn't necessarily become public.)
But the outcome of this litigation is hard to predict. Numerous lawsuits have been filed against AI companies, alleging that tools such as Stable Diffusion and ChatGPT easily copy copyrighted works and illegally train on them at scale. The companies counter that such copies are rare and difficult to make, and that the training should be considered fair use in law. Some of the earlier cases have been dismissed, such as the GitHub Copilot lawsuit that was dismissed in yesterday's ruling. Others, such as the New York Times Company's lawsuit against OpenAI, are ongoing.
At the same time, OpenAI, Google, and other tech giants are signing multimillion-dollar deals with publishers (including Vox Media, the parent company of The Verge) and photo providers for ongoing data access. The legal stakes are particularly high for both sides in this dispute, as smaller companies like Stability and Midjourney have less capital to buy data access, and individual artists have less leverage to demand payment.
