Andersen v. Stability AI is a class-action copyright lawsuit brought by visual artists against the makers of Stable Diffusion and related AI image generators, alleging their copyrighted artwork was scraped from the internet and used to train the models without permission. Filed in January 2023 in the U.S. District Court for the Northern District of California, the case is in discovery before Judge William H. Orrick, with trial scheduled for September 8, 2026. It has not settled.1
Who Is Suing
The suit was originally filed by three artists: Sarah Andersen, Kelly McKernan, and Karla Ortiz. An amended complaint added seven more plaintiffs: Grzegorz Rutkowski, Gregory Manchess, Gerald Brom, Jingna Zhang, Julia Kaye, Adam Ellis, and Hawke Southworth.
Their core allegation is that the defendants scraped billions of images from the internet, compiled them into the LAION-5B and LAION-400M datasets, and used those datasets to train Stable Diffusion without consent, credit, or compensation. The plaintiffs contend the trained model contains compressed or algorithmic representations of their work and can produce output “in the style of” specific artists whose images were included in training.
Beyond direct copyright infringement, the plaintiffs have pleaded induced infringement, arguing the defendants enabled end users to generate infringing images by distributing Stable Diffusion. Against Midjourney, they added a false endorsement claim under the Lanham Act, tied to Midjourney’s alleged publication of lists of artists whose styles could be replicated and a showcase site displaying user-generated images. Trade-dress claims target the capture of the artists’ distinctive visual identity.
The plaintiffs are represented by the Joseph Saveri Law Firm, attorney Matthew Butterick, and co-counsel Lockridge Grindal Nauen P.L.L.P. The defense team includes attorney Mark Lemley, who entered an appearance early in the case.
Who the Defendants Are
The original defendants are Stability AI Ltd. and Stability AI, Inc., which built Stable Diffusion; DeviantArt, Inc., which offered an AI tool called DreamUp built on Stable Diffusion; and Midjourney, Inc., which runs its own image-generation platform. Runway AI, Inc. was added in the first amended complaint filed in November 2023, on allegations that it worked with Stability AI to train and distribute versions of Stable Diffusion and offered a text-to-image tool through its “AI Magic Tools” product.
Stability AI is the primary target as the alleged scraper and dataset builder. Midjourney and DeviantArt are alleged to have incorporated Stable Diffusion into consumer-facing products. In early motion practice, the court observed that DeviantArt essentially provided its users “access to Stable Diffusion as a library” and that the plaintiffs had not adequately shown those companies played a direct role in scraping copyrighted works.
Key Rulings So Far
The case has moved through several rounds of motions to dismiss and amended complaints. Some claims have been narrowed or thrown out; the core copyright theories are heading to trial.
October 2023 Dismissal
Judge Orrick granted the defendants’ first motions to dismiss on nearly every claim, with one exception: Sarah Andersen’s direct copyright infringement claim against Stability AI survived. The court said the remaining claims lacked “specific plausible facts” and gave the plaintiffs leave to amend.
August 2024 Ruling on the Amended Complaint
After the plaintiffs filed a first amended complaint adding parties, plaintiffs, and revised theories, the defendants moved to dismiss again. On August 12, 2024, Judge Orrick granted those motions in part and denied them in part.
The court found it plausible that Stable Diffusion contains copyrighted works in some form and allowed two direct infringement theories to proceed. The “model theory” treats the AI product itself as an infringing copy because it embodies transformations of protected works. The “distribution theory” treats distributing the AI product as equivalent to distributing the copyrighted works it contains. Induced infringement claims also survived, with the court noting “plausible inferences at this juncture are that Stable Diffusion by operation by end users creates copyright infringement and was created to facilitate that infringement by design.”
Judge Orrick rejected the defendants’ comparison of AI image generators to VCRs, which the Supreme Court found noninfringing. Unlike VCRs, he reasoned, AI models like Stable Diffusion are “built to a significant extent on copyrighted works” and may be designed to “invoke copies or protected elements.” He also distinguished this case from Kadrey v. Meta, a lawsuit involving large language models, holding that image generators are “materially different” from text-based AI systems and that precedents from one type of AI model may not carry over.
Some claims were dismissed permanently. All Digital Millennium Copyright Act claims were thrown out with prejudice. The court held that because no plaintiff had shown an AI-generated output identical to one of their works, there could be no claim for removal of copyright management information under existing Ninth Circuit precedent. Unjust enrichment claims were dismissed as preempted by the Copyright Act.
Later Amendments
The plaintiffs filed a second amended complaint in October 2024 and a third amended complaint on February 27, 2026. Defendants filed their answers to the third amended complaint by March 13, 2026. The specific new theories in the latest complaint are not publicly detailed in available court records.
What the Trial Will Turn On
Fair use is expected to be the central defense. The defendants’ position rests on cases holding that copying is generally noninfringing when it serves a “nonexpressive” purpose, such as extracting metadata or enabling indexing. Key precedents include Authors Guild v. Google and Authors Guild v. HathiTrust, both of which approved large-scale digitization of copyrighted books for search and research.
Applied to AI training, the argument is that models like Stable Diffusion do not store or reproduce copyrighted images but learn abstract patterns, features, and statistical associations from the training data. That defense faces complications when models “memorize” training images. Academic research has shown that training images can sometimes be reproduced through precise prompts, a point Judge Orrick highlighted in his August 2024 ruling, along with statements from Stability AI’s former CEO about the model’s ability to “recreate” images from its dataset.
Other AI training rulings have cut in the defendants’ direction, but in different contexts. In Bartz v. Anthropic, a federal judge found in June 2025 that using legally acquired books for LLM training was “exceedingly” transformative and thus fair use, while obtaining those books through piracy was not protected. In Kadrey v. Meta, another court found AI training “highly transformative.” Whether those text-model rulings will influence outcomes for image generators is unresolved, and Judge Orrick has been explicit that different AI technologies warrant different legal analysis.
The court has flagged that several key questions, including whether Stability AI’s training process qualifies as fair use and whether the model and distribution theories hold up under factual scrutiny, are “better addressed on summary judgment, after discovery.”
The Parallel Getty Ruling in the UK
A separate lawsuit brought by Getty Images against Stability AI in the United Kingdom reached judgment on November 4, 2025. Getty had brought claims for copyright infringement, trademark infringement, database rights infringement, and passing off, but abandoned its primary copyright infringement claims before closing submissions, conceding there was no evidence Stability AI’s model training had occurred in the UK.
On the remaining claims, the English High Court found limited trademark infringement tied to Getty’s watermarks appearing in some AI outputs but rejected Getty’s secondary copyright infringement claim. Justice Joanna Smith held that the Stable Diffusion model “does not store any of those Copyright Works” and that “the model weights are not themselves an infringing copy.” Getty has said it intends to use findings of fact from the UK ruling in its ongoing U.S. litigation against Stability AI.
The ruling cuts both ways for Andersen. Getty’s failure on secondary copyright infringement supports Stability AI’s position that models do not “contain” copyrighted works. But the UK court’s finding that copyrighted works were in fact used to train Stable Diffusion, and its conclusion that intangible articles like AI models can be subject to copyright infringement claims just as tangible ones can, could bolster the plaintiffs’ arguments in California.
What’s Next Before September 2026
The case is in discovery. In a joint case management statement filed on January 13, 2026, the plaintiffs said they anticipate filing a motion for class certification; the defendants said they will oppose. No class certification motion has been filed yet, and no summary judgment motions have been filed. Trial is set for September 8, 2026.
Stability AI has signaled interest in working with creators outside the courtroom. CEO Prem Akkaraju has said the company is developing a marketplace or portal for artists to opt in to having their work used for AI training, with licensing and compensation, though he acknowledged “the economics of how to pay for that content hasn’t been ironed out yet.” Public filings and statements to date give no indication the parties are moving toward settlement.