AI Image Generator Can’t Escape Artists’ Copyright Class Action Lawsuit Over AI Training

A federal judge in San Francisco ruled Monday (Oct. 30) that artificial intelligence (AI) firm Stability AI could not dismiss a lawsuit claiming it had “trained” its platform on copyrighted images, though he also sided with AI companies on key questions.

In an early-stage order in a closely watched case, Judge William Orrick found many defects in the lawsuit’s allegations, and he dismissed some of the case’s claims. But he allowed the case to move forward on its core allegation: That Stability AI built its tools by exploiting vast numbers of copyrighted works.

“Plaintiffs have adequately alleged direct infringement based on the allegations that Stability downloaded or otherwise acquired copies of billions of copyrighted images without permission to create Stable Diffusion, and used those images to train Stable Diffusion,” the judge wrote.

The ruling came in one of many cases filed against AI companies over how they use copyrighted content to train their models. Authors, comedians and visual artists have all filed lawsuits against companies including Microsoft, Meta and OpenAI, alleging that such unauthorized use by the fast-growing industry amounts to a massive violation of copyright law.

Last week, Universal Music Group and others filed the first such case involving music, arguing that Anthropic PBC was infringing copyrights en masse by using “vast amounts” of music to teach its software how to spit out new lyrics.

Rulings in the earlier AI copyright cases could provide important guidance on how such legal questions will be handled by courts, potentially impacting how UMG’s lawsuit and others like it play out in the future.

Monday’s decision came in a class action filed by artists Sarah Andersen, Kelly McKernan and Karla Ortiz against Stability AI Ltd. over its Stable Diffusion — an AI-powered image generator. The lawsuit also targeted Midjourney Inc. and DeviantArt Inc., two companies that use Stable Diffusion as the basis for their own image generators.

In his ruling, Judge Orrick dismissed many of the lawsuit’s claims. He booted McKernan and Ortiz from the case entirely and ordered the plaintiffs to re-file an amended version of their case with much more detail about the specific allegations against Midjourney and DeviantArt.

The judge also cast doubt on the allegation that every “output” image produced by Stable Diffusion would itself be a copyright-infringing “derivative” of the images that were used to train the model — a ruling that could dramatically limit the scope of the lawsuit. The judge suggested that such images might only be infringing if they themselves looked “substantially similar” to a particular training image.

But Judge Orrick included no such critiques for the central accusation that Stability AI infringed Andersen’s copyrights by using them for training without permission — the basic allegation at the center of all of the AI copyright lawsuits, including the one filed by UMG. Andersen will still need to prove that such an accusation is true in future litigation, but the judge said she should be given the chance to do so.

“Even Stability recognizes that determination of the truth of these allegations — whether copying in violation of the Copyright Act occurred in the context of training Stable Diffusion or occurs when Stable Diffusion is run — cannot be resolved at this juncture,” Orrick wrote in his decision.

Attorneys for Stability AI, Midjourney and DeviantArt did not return requests for comment. Attorneys for the artists praised the judge for allowing their “core claim” to move forward and onto “a path to trial.”

“As is common in a complex case, Judge Orrick granted the plaintiffs permission to amend most of their other claims,” said plaintiffs’ attorneys Joseph Saveri and Matthew Butterick after the ruling. “We’re confident that we can address the court’s concerns.”