To receive industry-leading AI updates and exclusive content, sign up for our daily and weekly newsletters. Learn more
Visual artists who have joined a class action lawsuit against some of the most popular AI image and video generation companies are in a celebratory mood today after a judge ruled that copyright infringement lawsuits against AI companies can move forward with discovery.
Disclosure: VentureBeat regularly uses an AI art generator to create artwork for our articles, including the ones named in this story.
The lawsuit, docketed under number 3:23-cv-00201-WHO, was originally filed in January 2023. It has since been amended several times and dismissed in part, including today.
Which artists are featured?
Artists Sarah Andersen, Kelly McKernan, Carla Ortiz, Hawk Southworth, Grzegorz Rutkowski, Gregory Manches, Gerald Blom, Jinnah Chang, Julia Kay, and Adam Ellis, all of whom are representing themselves, have accused Midjourney, Runway, Stability AI, and DeviantArt of copying their work by offering an AI image generation product based on the open-source Stable Diffusion AI model jointly developed by Runway and Stability AI. The artists allege that the model was trained on their copyrighted works in violation of the law.
The judge's ruling today
Judge William H. Orrick of the Northern District of California, which sits in San Francisco and the epicenter of the generative AI boom, has yet to rule on the case's ultimate outcome, but wrote in a ruling today that “the allegations of inducement of infringement are sufficient” for the case to move to the discovery phase, where the artists' lawyers will be able to peer inside the AI image-generating companies and examine internal documents that could reveal details about their training datasets, mechanisms, and inner workings to the world.
“This is a case in which Plaintiffs allege that Stable Diffusion makes substantial use of copyrighted works and that the manner in which its products operate necessarily invokes copies or protected elements of those works,” Orrick's ruling states. “Whether that is true and whether it was the result of a defect (Stability's allegation) or was intentional (Plaintiffs' allegation) remains to be tested. An allegation of induced infringement is sufficient.”
The artists responded with applause.
“The judge ruled in favor of our copyright claims and we can now find out in discovery everything these companies don't want us to know,” Kelly McKernan, one of the artists who filed the lawsuit, wrote on her account on social network X. “This is a huge win for us and I'm so proud of our incredible team of lawyers and fellow plaintiffs!”
Very exciting news regarding the AI lawsuits. The judge ruled in favor of our copyright claims, and discovery will now allow us to learn all the things these companies don't want us to know. This is a huge win for us, and I'm very proud of our incredible team of lawyers and fellow plaintiffs. pic.twitter.com/jD6BjGWMoQ
— Kelly McKernan (@Kelly_McKernan) August 12, 2024
“Not only will this order allow copyright infringement claims to go forward, but it also means that companies that use the SD (Stable Diffusion) model and datasets like LAION can be held liable for copyright infringement and other infringing activities,” Carla Ortiz, another plaintiff artist in the lawsuit, wrote on her X account.
1/3 Big update on our case!
This is a major win for us as the judge allowed all of our copyright infringement claims and allowed us to proceed with our Lanham Act (trade dress) claims as normal, so we can now move forward with discovery.
The impact of this order is enormous in many ways! pic.twitter.com/ZcoeFtPtQb
— Carla Ortiz (@kortizart) August 12, 2024
Technical and legal background
Stable Diffusion was reportedly trained on LAION-5B, a dataset of more than 5 billion images that researchers collected from across the web and posted online in 2022.
But as the lawsuit itself points out, the database only contained URLs or links to the images and text descriptions, meaning the AI companies would have had to separately scrape or screenshot copies of the images in order to train Stable Diffusion and other derivative AI model products.
A ray of hope for AI companies?
Orrick handed the AI image-generation companies a victory by dismissing and tossing lawsuits brought by artists against them under the Digital Millennium Copyright Act of 1998, which prohibits companies from offering products designed to circumvent restrictions on copyrighted material delivered online or through software (also known as “digital rights management,” or DRM).
Midjourney tried to reference older court cases “dealing with jewelry, wooden cutouts, and key chains,” in which it was decided that the similarity of a different jewelry product to a previous artist's jewelry product did not amount to copyright infringement because the similarity was a “functional” element that was necessary to display certain features or elements of real life, or what the artist was trying to create, regardless of the similarity to the previous work.
The artists argued that “the stable diffusion model uses 'CLIP-guided diffusion,' which relies on prompts, including the artist's name, to generate images.”
CLIP stands for “Contrastive Language-Image Pre-training” and is a neural network and AI training technique developed by OpenAI in 2021, more than a year before ChatGPT was released, that can identify objects in images and label them with natural language text captions, making it extremely useful for compiling datasets for training new AI models such as Stable Diffusion.
“Plaintiffs allege that the CLIP Model functions as a trade dress database that allows them to recall and reproduce elements of each artist's trade dress,” Orrick wrote in part of his ruling regarding Midjourney, adding that “the combination of identified elements and images adequately illustrates and plausibly supports Plaintiffs' trade dress claims, given Plaintiffs' assertions as to how the CLIP Model functions as a trade dress database, and Midjourney's use of Plaintiffs' names on the Midjourney Name List and in its Showcases.”
In other words, the fact that Midjourney used artists’ names and labeled elements of their works to train its models may constitute copyright infringement.
But as I've argued before, from my perspective as a journalist, not a copyright lawyer or an expert in the field, it is already possible, and legally permissible, to commission a human artist to create a new work in the style of a copyrighted artist's work, which would seem to weaken the plaintiffs' argument.
As the lawsuit progresses, we'll see how well AI Art Generator defends its training methods and model output. Read the full case embedded below.
VB Daily
Stay up to date! Get the latest news every day by email
Thanks for subscribing! Check out other VB newsletters here.
An error has occurred.
