Some artists have begun a legal battle against the alleged theft of billions of copyrighted images used to train AI art generators and reproduce unique styles without compensating artists or seeking permission.
A group of artists represented by Joseph Saveri Law Firm has filed a U.S. federal class action lawsuit in San Francisco against AI art companies Stability AI, Midjourney and DeviantArt alleging violations of the Digital Millennium Copyright Act, violations of the right of publicity and unlawful competition .
The artists taking action — Sarah Andersen, Kelly McKernan, Karla Ortiz — are “trying to put an end to this egregious and massive infringement of their rights before their profession is eliminated by a computer program powered entirely by their hard work,” according to the official text of the complaint submitted to the court.
Using tools like Stability AI’s Stable Diffusion, Midjourney, or the DreamUp generator on DeviantArt, people can type sentences to create works of art that resemble living artists. Since the mainstream rise of AI image synthesis in the past year, AI-generated artworks have been highly controversial among artists, sparking protests and culture wars on social media.
A notable absence from the list of companies named in the complaint is OpenAI, creator of the DALL-E image synthesis model that arguably set the ball rolling in mainstream generative AI art in April 2022. Unlike Stability AI, OpenAI has the exact contents of its training dataset and has commercially licensed some of its training data to companies such as Shutterstock.
Despite the Stable Diffusion controversy, the legality of how AI image generators work has not been tested in court, although the Joesph Saveri Law Firm is no stranger to legal action against generative AI. In November 2022, the same company filed a lawsuit against GitHub over its Copilot AI programming tool alleging copyright violations.
Weak arguments, ethical violations
Alex Champandard, an AI analyst who did advocated artist rights without outright rejecting AI technology, criticized the new lawsuit in several threads on Twitter, write, “I don’t trust the lawyers who filed this complaint, based on content + how it’s written. The case could do more harm than good because of this.” Still, Champandard thinks the lawsuit could be detrimental to the potential suspects: “Everything the companies say to theI myself will be used against them.”
As for Champandard, we’ve noticed that the complaint contains several statements that may misrepresent how AI technology for image synthesis works. For example, the fourth paragraph of Section I states, “When Stable Diffusion is used to produce images based on user instructions, Stable Diffusion uses the training images to produce what appear to be new images through a mathematical software process. These ‘new’ images are completely based on the Training Images and its derivative works of the specific images that Stable Diffusion draws from when compiling a given output. Ultimately, it is just a complex collage tool.”
In another section attempting to describe how latent diffusion image synthesis works, the plaintiffs falsely compare the trained AI model to “having a folder on your computer containing billions of JPEG image files,” claiming that “a trained diffusion model can produce a copy from one of the training images.”
During the training process, Stable Diffusion drew from a large library of millions of scraped images. Using this data, the neural network statistically “learned” how certain image styles appear without storing exact copies of the images it saw. Although in the rare cases of over-represented images in the dataset (such as the Mona Lisa), a type of “overfittingcan occur allowing Stable Diffusion to spit out an accurate representation of the original image.
Ultimately, latent diffusion models, when properly trained, always generate new images and do not create collages or duplicate existing work — a technical reality that may undermine the copyright infringement plaintiffs’ argument, even though their arguments about “derivative works” created by the AI image generators is an open question with no clear legal precedent to our knowledge.
Some other points of the complaint, such as unlawful competition (by duplicating an artist’s style and using a machine to replicate it) and infringement of the right of publicity (by allowing people to “use artwork” without permission the style” of existing artists), are less technical and may have legs in the courtroom.
Despite the issues, the lawsuit comes after a wave of anger over the lack of consent from artists who feel threatened by AI art generators. In their own words, the technology companies behind AI image synthesis acquired intellectual property to train their models without the consent of artists. They are already on trial in the court of public opinion, even if they are eventually found to be in line with established case law on the excessive collection of public data from the Internet.
“Companies building large models that rely on copyrighted data can get away with it if they do it privately,” tweeted Champandard, “but doing it openly *and* legally is very difficult – or impossible.”
Should the lawsuit go to trial, the courts will have to resolve the differences between ethical and alleged legal violations. The plaintiffs hope to prove that AI companies benefit commercially and profit lavishly from the use of copyrighted images; they have asked for substantial damages and a permanent injunction to prevent companies allegedly infringing from further violations.
When reached for comment, Emad Mostaque, CEO of Stability AI, replied that the company had not received any information about the lawsuit at the time of writing.
