Artists across the United States have filed copyright lawsuits against AI image generators including Stable Diffusion, Midjourney, and DeviantArt’s generative tools, alleging the companies trained their models on billions of copyrighted images scraped from the internet without permission or payment. The leading case, Andersen v. Stability AI, is in active discovery with trial scheduled for September 8, 2026. Getty Images, Disney, and Warner Brothers have since brought their own suits, dramatically expanding the stakes.
What Artists Are Claiming
The claims break into three groups. The first and largest is straightforward copyright infringement: when an AI company downloaded an image to include in a training set, plaintiffs say, that download was an unauthorized reproduction of a protected work. Federal copyright law reserves the right to reproduce a work, prepare derivative works from it, and control its distribution to the creator alone.1Office of the Law Revision Counsel. 17 USC 106 – Exclusive Rights in Copyrighted Works Multiply one unauthorized copy by the billions of images in a modern training set and the exposure becomes serious: statutory damages run from $750 to $30,000 per work, rising to $150,000 per work for willful infringement.2Office of the Law Revision Counsel. 17 USC 504 – Remedies for Infringement: Damages and Profits
Plaintiffs also argue the trained model itself is an infringing derivative work. On this theory, because every parameter inside the model was shaped by processing copyrighted art, the resulting software is a transformed version of the training library, and only the original creators can authorize works that incorporate their protected material. If the courts accept it, liability would extend not just to the initial scraping but to every copy of the model distributed afterward.
The second group of claims targets the Digital Millennium Copyright Act. Federal law prohibits anyone from intentionally removing copyright management information — the author’s name, copyright notices, embedded watermarks, and similar identifiers — when the removal is intended to enable or conceal infringement.3Office of the Law Revision Counsel. 17 USC 1202 – Integrity of Copyright Management Information Artists allege the companies systematically strip this metadata during the data preparation pipeline, severing the link between each image and the person who made it. Statutory damages here run from $2,500 to $25,000 per removal, applied separately from the underlying copyright penalties.4Office of the Law Revision Counsel. 17 U.S. Code 1203 – Civil Remedies
The third group concerns identity. The right of publicity, recognized in a majority of states, prevents commercial exploitation of a person’s name or likeness without consent. When an AI platform lets a paying user type “in the style of” followed by a specific artist’s name, plaintiffs argue, that artist’s creative identity has been packaged as a product feature. Some plaintiffs have added Lanham Act claims, which make it unlawful to use a name or designation in commerce in a way likely to confuse consumers about who created or endorsed a product.5Office of the Law Revision Counsel. 15 U.S. Code 1125 – False Designations of Origin, False Descriptions, and Dilution Forbidden Judges have narrowed some of the identity claims, and Lanham Act arguments require proof of actual consumer confusion that is hard to produce on a motion to dismiss, but these theories remain live alongside the copyright counts.
The Fair Use Fight
Fair use is the defense the AI companies are counting on. Federal law allows unauthorized copying in certain circumstances, and courts weigh four factors together: the purpose and character of the use, the nature of the work copied, the amount taken, and the effect on the market for the original.6Office of the Law Revision Counsel. 17 USC 107 – Limitations on Exclusive Rights: Fair Use
The defendants argue training a model to learn visual patterns is transformative, serving a fundamentally different purpose than displaying the artwork, and they point to the Supreme Court’s reasoning in Google LLC v. Oracle America that reimplementing copied material for a new platform can qualify as transformative.7Supreme Court of the United States. Google LLC v. Oracle America, Inc. The artists’ answer is that the models produce images directly competing with the works that trained them, which is not what Google was doing with Oracle’s code.
The remaining factors tilt against the defendants on their face. The training sets are packed with highly creative work, which gets stronger protection than factual compilations. Entire images were copied, not excerpts. And on market harm, a federal court recently granted summary judgment against an AI legal research tool partly on the ground that even non-identical AI output competed in the same market as the training data and displaced potential licensing revenue.8U.S. District Court for the District of Delaware. Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc. That ruling involved a legal research product, not an image generator, but the market-displacement logic maps onto the artist cases directly.
The U.S. Copyright Office took up the same question in its May 2025 report on generative AI training. The Office declined to prejudge the pending lawsuits but concluded the copying involved in AI training “threatens significant potential harm to the market for or value of copyrighted works,” added that outputs “can dilute the market for works similar to those found in [the] training data” even without reproducing any specific image, and said the availability of a licensing market cuts against fair use.9U.S. Copyright Office. Copyright and Artificial Intelligence, Part 3: Generative AI Training The report also flagged training on pirated or illegally obtained datasets as a factor weighing against fair use.
The Major Cases and Who’s Being Sued
The lead case is Andersen v. Stability AI, filed in January 2023 in the Northern District of California by visual artists Sarah Andersen, Kelly McKernan, and Karla Ortiz. The original defendants are Stability AI, which created and released Stable Diffusion as an open-source model; Midjourney, which runs a closed subscription image generator; and DeviantArt, which built generative tools into its artist portfolio platform.10CourtListener. Andersen v. Stability AI Ltd. The complaint pleads copyright infringement, DMCA violations, and right of publicity claims.
Stability AI drew particular attention because its open-source release put the trained model weights into thousands of third-party applications. Even if the company itself stopped operating tomorrow, the model would continue to spread. DeviantArt’s inclusion stung because the site had long functioned as a place for artists to share and sell original work.
The field widened sharply through 2025. Getty Images filed its own federal suit against Stability AI in the Northern District of California, alleging Stable Diffusion was trained on millions of Getty’s licensed photographs.11CourtListener. Getty Images (US), Inc. v. Stability AI, Ltd. Disney and Warner Brothers each sued Midjourney in the Central District of California. These corporate plaintiffs bring deep catalogs of properly registered works, which is the combination that makes large statutory damages claims credible.
One boundary worth noting: Getty pursued a parallel case in England, where the High Court issued a November 2025 judgment largely rejecting Getty’s claims, finding that Stable Diffusion does not store or reproduce copyrighted images internally and that there are “no copies in the model.” That ruling applies UK copyright law and has no binding effect on U.S. courts, though the defendants will cite it as persuasive.
Where the U.S. Cases Stand in 2026
The Andersen case has survived multiple rounds of motions to dismiss. In August 2024, the court allowed the core direct copyright infringement claim to proceed while narrowing or dismissing some of the broader theories.12Justia. Andersen et al v. Stability AI Ltd. et al The plaintiffs filed a third amended complaint in February 2026, the defendants answered in March 2026, and discovery is active. The artists can now demand internal documents, training logs, and company communications. Trial is set for September 8, 2026.
The Getty U.S. case is on a similar track. The court ruled on a motion to dismiss in April 2026, granting it in part and denying it in part, and referred the case to a magistrate judge for discovery.11CourtListener. Getty Images (US), Inc. v. Stability AI, Ltd. A jury trial is scheduled for January 2028.
Underneath all of these cases runs one disputed factual question: whether a trained AI model actually stores recognizable versions of copyrighted works or only learns statistical patterns. The defendants say no images exist inside the model and that generation reflects learned relationships between visual elements and text descriptions. The plaintiffs point to instances where models reproduce images strikingly similar to specific training examples and argue that proves the works are stored in some compressed form. Discovery is likely to turn on technical evidence that could resolve this before trial.
Why Registration Matters If You Want to Sue
If you are an artist considering joining this fight, start with the Copyright Office. You cannot recover statutory damages or attorney’s fees for copyright infringement unless you registered your work with the Copyright Office before the infringement began, or within three months of first publication.13Office of the Law Revision Counsel. 17 USC 412 – Registration as Prerequisite to Certain Remedies for Infringement Copyright attaches the moment you create the work, but the ability to collect meaningful damages depends on timely registration. Many digital artists who post hundreds of works online have never registered any of them.
Without registration, you can still sue for actual damages, but those are hard to quantify when a single image sits inside a dataset of billions. Statutory damages solve that problem by assigning a fixed per-work range, which is why the registration question is often decisive.
The Copyright Office offers group registration options for two-dimensional artwork and for unpublished works, letting artists register batches in a single application.14U.S. Copyright Office. Register Your Work: Registration Portal Artists with smaller individual claims also have the Copyright Claims Board, a federal small-claims tribunal with total damages capped at $30,000 per proceeding and statutory damages capped at $15,000 per infringed work.15Copyright Claims Board. Frequently Asked Questions
Defensive Tools Artists Are Using Now
Litigation is slow, so artists have turned to technical defenses. Two of the most widely adopted come from the University of Chicago’s Glaze Project. Glaze applies invisible alterations to an image that prevent AI models from accurately learning the artist’s style. Nightshade goes further, introducing changes that actively corrupt a model’s understanding of what it is looking at, so a model trained on treated images learns wrong associations between objects and visual patterns. Both are designed to be nearly invisible to human viewers.
On the infrastructure side, several opt-out signals now exist. Website operators can use robots.txt directives to block known AI crawlers. DeviantArt introduced “noai” and “noimageai” meta tags in 2022, signaling that content should not be used for AI training. The TDMRep protocol allows site owners to declare text and data mining restrictions at the site, page, or individual URL level. Whether AI companies honor these signals is itself contested, and the question feeds back into the lawsuits: ignoring an explicit opt-out undercuts a fair use defense.
The Content Authenticity Initiative has developed Content Credentials, a system built on C2PA technical standards that embeds cryptographically signed metadata into digital files, creating a tamper-evident record of who created the work, when, and with what tools. Wider adoption would make it harder for AI companies to claim they did not know a training image was copyrighted or who made it, weakening the deniability that currently complicates DMCA enforcement.