RSS GitHub
The Ledger A sourced historical record of AI

Artists File First Class Action Against AI Image Generators in Andersen v. Stability AI

A ledger entry in the culture archive, dated 2023-01-13.

Major
References

Summary

On January 13, 2023, the illustrators Sarah Andersen, Kelly McKernan and Karla Ortiz filed a proposed class action in federal court in San Francisco against Stability AI, Midjourney and DeviantArt. The complaint alleged that Stable Diffusion and the image generators built on it were trained on billions of images copied without the artists' consent, and that the tools could produce work "in the style" of named artists that competed with them. It was the first class action brought by visual artists against generative AI companies.

What Happened

The complaint was filed in the US District Court for the Northern District of California as case No. 3:23-cv-00201 by the Joseph Saveri Law Firm and the lawyer and designer Matthew Butterick, who in November 2022 had brought a similar suit over GitHub Copilot. The three named plaintiffs sued on behalf of a proposed class of artists whose work appeared in the training data. Stability AI was sued as the developer of Stable Diffusion, Midjourney for its image generator, and DeviantArt for DreamUp, a paid tool built on Stable Diffusion.

The plaintiffs' central claim was that Stable Diffusion was trained on about five billion images scraped from the web, including the LAION-Aesthetics dataset, without the knowledge, consent or payment of the artists. The complaint cited an estimate that about 3.3 million images in that dataset came from DeviantArt. The lawyers' public summary described Stable Diffusion as "a 21st-century collage tool" that stores compressed copies of its training images and recombines them. Beyond direct copyright infringement, the complaint brought claims under the Digital Millennium Copyright Act, the right of publicity, California unfair competition law and, against DeviantArt, breach of its terms of service. It alleged that letting users request images "in the style" of a named artist used the artists' names to sell their style and siphoned commissions from them. The plaintiffs asked for damages and an injunction requiring changes to the image generators.

The case narrowed and then advanced. In October 2023 Judge William Orrick dismissed most of the original claims with leave to amend. The plaintiffs refiled in November 2023, joined by seven more artists, and Runway AI was also a defendant by the next ruling. On August 12, 2024, Judge Orrick allowed the copyright and trademark claims against Stability AI, Midjourney, DeviantArt and Runway AI to proceed, while dismissing the unjust enrichment, breach of contract and Digital Millennium Copyright Act claims. That ruling did not decide whether training on the works was fair use.

Why It Matters

This suit began the visual artists' side of the AI copyright dispute, eleven months before The New York Times sued OpenAI. Suits by Getty Images, authors, news publishers and music labels followed over the next two years. The complaint set out the questions those later cases kept returning to: whether copying works into a training set is infringement, whether a trained model contains copies of its training data, and whether output "in the style" of an artist harms the market for that artist's work.

The early dismissal of most claims showed that the "collage" theory of how diffusion models work would be contested on the technical facts, and the case moved forward mainly on narrower claims about copies stored in and reproduced by the models. Whether fair use protects training, the issue artists cared about most, was left for later stages and for other courts.

§ How to read the metadata
Landmark
Fundamentally alters the trajectory; 2–5 per year.
Major
Meaningfully shifts the landscape; 2–4 per month.
Notable
Worth documenting; significance can be upgraded later.
Confidence
High = primary sources corroborate. Medium = credible secondary only. Low = provisional. Disputed = credible sources disagree.
Contestation
Uncontested = no formal challenge. Contested = at least one challenge filed and not resolved by a correction. Corrected = a challenge was upheld and the entry corrected; see its revision history. Superseded = replaced by a later entry. Unresolved = dispute still open.

References

  1. Andersen v. Stability AI Ltd., No. 3:23-cv-00201 (N.D. Cal.), docket , CourtListener (Free Law Project) () primary document archived copy
  2. Stable Diffusion litigation , Joseph Saveri Law Firm and Matthew Butterick () primary document archived copy
  3. Artists and Illustrators Are Suing Three A.I. Art Generators for Scraping and 'Collaging' Their Work Without Consent , Artnet News () secondary reporting archived copy
  4. Artists score a victory in landmark AI copyright case , The Art Newspaper () secondary reporting archived copy

See also