Two companies that have developed artificial intelligence (AI) image generators argued for the dismissal of a class action lawsuit brought by artists who claim the businesses scraped their work without permission to train a text-to-image model.
The artists in the case allege that the AI software company Midjourney and the online art platform DeviantArt used their art to train a generative image tool, Stable Diffusion, without obtaining permission from the artists or paying them for their work, resulting in unfair marketplace competition.
The artists originally sued DeviantArt, Midjourney and Stable Diffusion creator Stability AI in January 2023, but much of the case was thrown out by US District Senior Judge William Orrick last October, allowing only a single direct copyright claim to stand against Stability AI. After the plaintiffs submitted an amended complaint, Judge Orrick made a tentative ruling on Tuesday (7 May) announcing his intention to allow direct and induced infringement claims against the defendants to proceed since the artists have plausibly claimed that copies of their work is stored in various versions of the tool, according to Courthouse News.
Orrick also ruled that the artists’ Digital Millennium Copyright Act claims should be tossed, since the plaintiffs had not shown that the creators of Stable Diffusion distributed or reproduced their works.
Andrew Gass, a lawyer for DeviantArt, argued in a hearing on 8 May for dismissal on the grounds that Stable Diffusion only used the works to produce images for its DreamUp AI tool, not to train an AI model.
“There is no allegation that DeviantArt itself ever extracted from the model any output that is a copyright infringement of the plaintiffs works. There's no allegation that any of DeviantArt users have ever extracted any output from DeviantArt’s implementation of the model,” Gass said, going on to describe DeviantArt’s actions as “classic fair use”, according to Courthouse News.
Laura Matson, a lawyer for the artists, argued that the images were still “reproducible”, claiming that it was inappropriate to discuss fair use at the dismissal stage.
Christopher Young, another representative for the artists, also drew a distinction between his clients’ case and a similar case in which comedian and author Sarah Silverman brought a lawsuit against Meta over its use of her books to train AI. The judge in that case ruled that the claims should be dismissed, since a model cannot infringe copyrighted works unless it readapts the original text. Young drew a contrast between text generators and image generators, insisting the two should not be compared, and underscored the volume of evidence his team had brought to the court.
Orrick took the statement under submission.