Three artists take AI image generators to court over copyright

For years, working artists put their portfolios online to build a career. Then came text-to-image models, and suddenly anyone could type an illustrator's name into a prompt and churn out near-identical artwork in seconds, with zero permission and zero royalties.

Kelly McKernan, an illustrator who uses watercolor and gouache for books and games, discovered this firsthand. A search on the site Have I Been Trained revealed their art was baked into Stable Diffusion, the open-source engine powering viral apps like Lensa.

Now, the creators are taking the tech industry to court.

The artists take on the machines

McKernan, joined by popular cartoonist Sarah Andersen and film concept artist Karla Ortiz, filed a class-action lawsuit in San Francisco against Stability AI, Midjourney, and DeviantArt.

Ortiz has designed for blockbusters like Jurassic World and Doctor Strange, while Andersen's signature comic style was scraped and co-opted on 4chan to make hateful memes. They felt violated, and they decided to fight back.

Representing them is Matthew Butterick and the Joseph Saveri Law Firm, the legal duo already suing Microsoft, GitHub, and OpenAI over Copilot. The suit claims AI models ingested billions of copyrighted images without consent to build commercial engines that directly displace human creators.

Days later, Getty Images also sued Stability AI in London, accusing the company of unlawfully scraping millions of stock photos along with their metadata.

Stability AI hit back quickly. A spokesperson claimed that anyone calling their process copyright infringement simply misunderstands both the technology and the law.

Collages, weights, and legal crosshairs

Tech founders have previously been remarkably candid about how these datasets were assembled. Midjourney CEO David Holz admitted to Forbes that their training data was essentially a massive scrape of the internet, noting it is impossible to trace rights across hundreds of millions of images.

Stability AI chief Emad Mostaque argued that fine art makes up less than 0.1 percent of the LAION-5B dataset. Yet as concept artist Jon Lam pointed out, even a fraction of a percent still represents millions of copyrighted images.

Stability AI tried offering an opt-out form for its next release, but artists pushed back immediately. Their stance is clear: consent should be opt-in from day one.

Still, legal scholars see flaws in the plaintiffs' argument. Jessica Fjeld, assistant director of the Cyberlaw Clinic at Harvard, noted that the suit mischaracterizes Stable Diffusion as a collage tool that stores compressed copies of source images.

In reality, diffusion models store mathematical weights and abstract relationships, not actual image files. Proving mechanical reproduction will be an uphill battle under existing copyright doctrine.

Why it matters

This legal clash marks the end of the free-for-all era in generative AI.

For independent artists, illustrators, and animators, the stakes are existential. If platforms can freely absorb an artist's signature visual identity and offer it as an instant style preset, the freelance creative economy faces massive devaluation.

For AI startups and enterprise developers, legal risk has officially arrived. While Silicon Valley long relied on the 2016 Google Books fair use precedent to justify web scraping, text-to-image engines produce direct market substitutes rather than search indexes.

We think the artists made a misstep by framing AI as a basic collage machine, which gives defense attorneys an easy technical target. However, filing this suit was the right move. It punctures the assumption that internet art is free real estate, and it will likely force the AI industry toward paid licensing deals whether it likes it or not.