Artists win important victory in copyright case against Stability AI and Midjourney

A group of artists who have joined together in a class-action lawsuit against the developers of some of the world’s most popular image-generating artificial intelligence models celebrated as a judge cleared the case and ordered disclosure.

Image source: Alexandra_Koch / pixabay.com

The defendants in the case are the creators of the services Midjourney, Runway, Stability AI and DeviantArt – according to the plaintiffs, the developers of systems based on the Stable Diffusion model used their copyrighted works to train AI. California Northern District Court Judge William H. Orrick, who presides over San Francisco, home to many of the world’s top AI developers, has not yet ruled on the case, but said the charges against the defendants were sufficient for the case to proceed to discovery. information. This means that lawyers representing plaintiffs can review documents from companies that develop AI image generators; Details about training data sets, mechanisms and internal workings of systems will be made public.

The Stable Diffusion model was supposedly trained on the LAION-5B dataset of 5 billion images, which was published in 2022. But, as noted in the case, this database contained only URLs, that is, links to images, as well as their text descriptions, meaning companies had to collect these images themselves.

Models based on Stable Diffusion use a CLIP-guided diffusion mechanism to help them generate images based on user requests, which may include artist names. The CLIP (Contrastive Language-Image Pre-training) method was developed and published by OpenAI back in 2021 – more than a year before the release of ChatGPT. The OpenAI CLIP model is capable of acting as a brand identity database, and if the similar Midjourney model was trained using artist names and their works with associated descriptions, this could constitute copyright infringement.