Hayao Miyazaki, Matt Groening and Keith Haring among artists allegedly used to train Midjourney AI

Amidst an ongoing class-action lawsuit between a group of artists and major AI companies, it has been revealed that Midjourney gathered a database of artists to train its text-to-image tool, allowing the easy reproduction of hugely popular artistic styles.

A leaked Google Sheet was circulating online on 31 December 2023 featuring visual styles like “xmaspunk”, “carpetpunk” and “polaroidcore” and thousands of individual artist names. The list was allegedly put together by developers at Midjourney; screenshots shared by Riot Games’ Jon Lam (founder of advocacy moment Create Don’t Scrape) appeared to show developers discussing sourcing artists and styles from Wikiart and laundering datasets to avoid copyright infringement.

The list spans huge names from history as well as some of the biggest creatives in visual culture today, with artists like William Blake and Henri Matisse gathered alongside the likes of Keith Haring, Matt Groening, Wes Anderson, Katsuhiro Otomo, Frida Kahlo, Hayao Miyazaki, Banksy, David Hockney, Trove Jansson, Anish Kapoor and Yayoi Kusama. Developers even discussed emulating the style of Bob Ross in internal conversations.

Not only leading names have been pooled by Midjourney – one X user (previously Twitter) pointed out that six-year-old Hyan Tran was on the database, who contributed artwork for a 2021 charity trading card edition of Magic: The Gathering raising money for Seattle Children’s Hospital Autism Center.

The original Google Sheet is no longer available to view but another version has been uploaded on the Internet Archive. It was first published as evidence in the ongoing legal battle between several artists and Midjourney, Stability AI and DeviantArt – a judge dismissed claims against Midjourney and DeviantArt in October. The document was taken from 455 pages of supplementary evidence filed in November.

Artists who have had their work scraped by AI models are facing uncertain legal territory when it comes to AI and copyright, especially as it becomes increasingly difficult to show substantial similarity between an artist’s work and AI outputs when models interpret elements of an artist’s style, like linework, rather than creating identical copies.