The legality of employing artists’ creations to train a generative AI to replicate their artistic style is at the heart of a current legal dispute between artificial intelligence experts and the artistic community in the United States.

This legal action specifically targets image generation systems offered by Stability, DeviantArt, Runway AI, and Midjourney. Midjourney, in particular, a widely popular platform among the general public, is under intense scrutiny based on documents submitted in the case. Internal communications among Midjourney developers reportedly revealed a list of 4,700 items, including artworks and styles, used to “train” the intelligence in creating images by emulating their composition and style.

The compilation of these lists, as per internal exchanges, originally stemmed from information sourced from Wikipedia and a comprehensive roster of artists who contributed to the Magic: The Gathering game series. The exchanges also indicated that developers were aware as early as 2022 of copyright issues associated with AI models leveraging existing artworks and artists.

“All you have to do is just use those scraped datasets and then conveniently forget what you used to train the AI model. Boom: legal problems solved forever,” reads a screenshot that was widely circulated in the American press.

Intellectual property laundering

The case goes even further as this approach currently leads AI to generate visuals solely based on an existing artist’s name. One of the plaintiffs, Kelly McKernan, was recently surprised to discover that the top online search result for her now showcases an artificial image replicating her artistic style.

“The AI image devices should primarily be seen as intellectual property laundering machines, promising users the benefits of art without the cost associated with engaging an artist,” stated the plaintiffs in the legal proceedings.

Initially filed by illustrators Sarah Andersen, Kelly McKernan, and Karla Ortiz, the collective lawsuit has since been joined by artists H. Southworth, Grzegorz Rutkowski, Gregory Manchess, Gerald Brom, Jingna Zhang, Julia Kaye, and Adam Ellis.

The original three plaintiffs were dismissed by a Californian federal court at the end of November in their case against Midjourney and DeviantArt. The decision was based on the challenge of proving copyright infringement unless the images are nearly identical. The court, however, left the door open for a revised complaint, highlighting the difficulty in proving that artists’ works had indeed been used to create AI-generated images.

This case, poised to set a legal precedent, is being closely monitored by the entire ecosystem of AI service providers. It is also being closely watched within the artistic realm, particularly in the fashion and luxury industries, where AI is perceived as both an additional tool and a potential threat by industry professionals.