A California judge has again changed the course of a keenly-followed case brought against developers of AI text-to-image generator tools by a group of artists, dismissing a number of the artists’ claims while allowing their core complaint of copyright violation to endure.

On August 12, Judge William H. Orrick, of the United States District Court of California, granted several appeals from Stability AI, Midjourney, DeviantArt, and a newly added defendant, Runway AI. This decision dismisses accusations that their technology variably violated the Digital Millennium Copyright Act, which intends to protect internet users from online theft; profited unfairly from the artists’ work (so-called “unjust enrichment”); and, in the case of DeviantArt, violated assumptions that parties will act in good faith towards contracts (the “covenant of good faith and fair dealing”). 

However, “the Copyright Act claims survive against Midjourney and the other defendants,” Orrick wrote, as do the claims regarding the Lanham Act, which protects the owners of trademarks. “Plaintiffs have plausible allegations showing why they believe their works were included in the [datasets]. And plaintiffs plausibly allege that the Midjourney product produces images—when their own names are used as prompts—that are similar to plaintiffs’ artistic works.”

In October of last year, Orrick dismissed a handful of allegations brought by the artists—Sarah Andersen, Kelly McKernan, and Karla Ortiz—against Midjourney and DeviantArt, but allowed the artists to file an amended complaint against the two companies, whose system utilizes Stability’s Stable Diffusion text-to-image software.

“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 October judgement.

In January 2023, Andersen, McKernan, and Ortiz filed a complaint that accused Stability of “scraping” 5 billion online images, including theirs, to train the dataset (known as LAION) in Stability Diffusion to generate its own images. Because their work was used to train the models, the complaint argued, the models are producing derivative works.

Midjourney claimed that “the evidence of their registration of newly identified copyrighted works is insufficient,” according to one filing. Instead, the works were “identified as being both copyrighted and included in the LAION datasets used to train the AI products are compilations.” Midjourney further contended that copyrighted protection only covers new material in compilations and alleged that the artists failed to identify which works within the AI-generated compilations are new.