Artists Suing Stability AI Forced to Go Back to The Drawing Board on Copyright Claims

A group of artists suing generative AI companies for allegedly using their copyrighted works are down, but not out, following a recent federal judge’s order. On Monday, the judge presiding over a case brought by three visual artists dismissed the majority of the claims levied against Stability AI, Midjourney, and art social network DeviantArt after determining the artists’ accusations were “defective in numerous respects.”

All of the allegations against Midjourney and DeviantArt were dismissed, though the artists and their legal teams will have a chance to amend their complaint to state their argument more clearly. The core question of whether or not training generative AI models on artists’ work amounts to a copyright infringement, however, remains totally unresolved.

Why did the artists go to court?

The case stems from a January lawsuit filed by artists Sarah Andersen, Kelly McKernan, and Karla Ortiz, who accused the tech companies of downloading billions of copyrighted images to train models without the artist’s consent and without compensating them. The artists claim supposedly “new” creations generated by Stability AI’s Stable Diffusion generator derivative of their own supposedly sucked up on a dataset used to train the models. Generative AI image generators, the artists argued in their lawsuits, aren’t creating completely original art, but are instead “merely a complex collage tool.” The artists sought a permanent injunction from the court barring Stability AI and the other defendants from using artwork without artists’ permission.

But the case isn’t going all according to plan for the artists so far. In an order filed this week, U.S. District Judge William Orrick dismissed the copyright infringement complaint brought by McKernan and Ortiz because neither of them registered their creations with the US Copyright Office. Andersen, best known as the author of the webcomic “Sarah Scribbles” has 16 registered collections she believes were used to train Stability’s AI models.

Orrick seemed unconvinced about whether or not the actual images generated by the AI models amount to a copyright infringement. In their complaint, the artists particularly took issue with AI images generated via prompts asking the modes to create images “in the style of” a known professional. The images the AI models spit out, they argue, end up competing in the marketplace against the original work of the human artist they were based on. But many of the works generated by these models, even if they are trained on an artist’s original work, may not look similar enough to the original artist’s work to run afoul of copyright protection. In other words, those “inspired by” images generated by AI models likely do not violate the artist’s copyright.

“I am not convinced that copyright claims based on a derivative theory can survive absent ‘substantial similarity’ type allegations,” Orrick wrote in the order. “The cases plaintiffs rely on appear to recognize that the alleged infringer’s derivative work must still bear some similarity to the original work or contain the protected elements of the original work.”

The judge also expressed skepticism towards the artists’ theory that the billions of supposedly scrapped works were “compressed” into Stable Diffusion’s program. Stability AI has previously denied accusations that training its AI model requires complete copies of copyrighted works. Instead, Stability claims it trains its models using complex parameters that are associated with certain subjects.

“Plaintiffs will be required to amend to clarify their theory with respect to compressed copies of Training Images and to state facts in support of how Stable Diffusion—a program that is open source, at least in part—operates with respect to the Training Images,” Orrick’s order stated.

Stability AI did not immediately respond to Gizmodo’s request for comment. Matthew Butterick, and Joseph Saveri, two of the attorneys representing the artists, told Gizmodo they believed the order issued by Judge Orrick sustained their client’s “core claim” regarding the alleged direct copyright infringement by Stability AI. That core claim, he said, was now on a path to trial.

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

AI firms face a flurry of copyright lawsuits

Variations of that “core” infringement claim Orrick allowed to proceed are at the heart of several other big-name lawsuits lodged against generative AI firms by authors, other artists, and record labels. Comedian Sarah Silverman joined two other authors in a lawsuit lodged against OpenAI and Meta earlier this year accusing the company of training its OpenAI and LLaMa large language models on their copyrighted materials. Subsequent analysis of the Books3 database reportedly used to help train both of those models includes text from more than 183,000 books.

More recently, record label Universal Music Group and several other major music publishers sued Anthropic for allegedly distributing copyright lyrics in its Clause 2 AI models. Like the artist and author cases, the music publishers’ complaint similarly accused Anthropic of using copyrighted material to train its language models.

Back on the visual side of things, stock image giant Getty Images sued Stability AI for allegedly “stealing” 12 million of its copyrighted images to train its Stable Diffusion model. Getty asked a UK court to force Stability AI to remove the supposedly violating images and pay them $150,000 for each “infringed image.” Not to be outdone, Getty turned heel last month and announced its own AI image generator trained entirely on its own images. In theory, Getty believes that reliance on in-house images should steer them clear of the same types of copyright claims they are forcing tech companies to endure.

Moving forward, the judge’s order in the artist’s case against Stability AI suggests that continuing copyright rights with AI companies could be a relatively narrow one.