AI Image Lawsuits: Getty, Midjourney, and Demand Letters

Lawsuits over images now move on two very different tracks. On one, photographers, artists, stock agencies, and movie studios are suing AI companies that trained image generators on their work. On the other, enforcement firms send thousands of demand letters a month to businesses and website owners who used a photo without a license. The AI cases will set the rules for a new industry; the demand-letter cases are already costing ordinary businesses thousands of dollars each. By mid-September 2025, at least fifty copyright lawsuits had been filed against AI companies over the use of copyrighted works in training data.¹

The AI Image Lawsuits: Who Is Suing Whom

The central question in nearly every AI image case is the same. Does feeding copyrighted images into a model without permission infringe copyright, or is it a transformative fair use because the output is a new technological tool?

AI developers lean on Supreme Court decisions like Google v. Oracle and Sony v. Universal, arguing that training a model is exactly the kind of innovation copyright is meant to encourage.¹ Rights holders answer that mass scraping without payment is free-riding on creators to build a product that competes with them.

Three cases stand out.

Getty Images v. Stability AI

Getty, one of the world’s largest stock photo agencies, sued Stability AI, the company behind Stable Diffusion. Getty’s original 2023 complaint alleged that Stability AI copied more than 12 million photographs along with their captions and metadata.² A striking piece of the case: Stable Diffusion sometimes generated images containing distorted Getty watermarks, which Getty argued misled viewers about the source.

Getty refiled in the Northern District of California in August 2025 after dismissing the original Delaware action.³ Judge Trina L. Thompson is presiding, with trial set for January 2028.⁴ On April 23, 2026, Judge Thompson allowed Getty’s claims for direct copyright infringement, trademark infringement, trademark dilution, and California unfair competition to proceed. She rejected Stability AI’s argument that the Supreme Court’s Dastar decision barred the trademark claims, reasoning that Stability AI was distributing its own AI-generated images in a way that could mislead consumers about source, not just repackaging an existing product.⁵

The court dismissed Getty’s DMCA claim for false copyright management information, finding Getty had not adequately alleged the specific intent the statute requires. The judge described the distorted watermarks as “incidental byproducts” of training rather than a deliberate effort to conceal infringement, and dismissed that claim without prejudice.⁵ Fact discovery closes in September 2026, and the parties are in court-ordered mediation with an October 2026 deadline.⁴

Getty also sued in England’s High Court. On November 4, 2025, Mrs. Justice Joanna Smith issued the first UK ruling on AI and copyright, and it was mostly a loss for Getty.⁶ The court dismissed the secondary copyright infringement claim, holding that Stable Diffusion’s model weights “do not store” the training images and are “not themselves an infringing copy.” Because Getty conceded at trial there was no evidence training occurred in the UK, it abandoned its primary copyright and database rights claims, so the court never ruled on whether scraping images to train a model is infringement under UK law.⁷

Getty did win a narrow trademark finding against older Stable Diffusion versions (v1 and v2) that produced images with Getty and iStock watermarks. The court called these findings “extremely limited in scope” and dismissed trademark claims against newer versions.⁸ Both sides are pursuing appeals; the trial judge granted Getty permission to appeal the copyright ruling, calling it a “novel and important” question of law. A Court of Appeal decision is expected by late 2026 or early 2027.⁹

Andersen v. Stability AI: The First AI Jury Trial

A class action by visual artists against Stability AI, Midjourney, and DeviantArt is set for jury trial on September 8, 2026, the first AI copyright case to reach a US jury.³ In August 2024, Judge William Orrick allowed the artists’ “model theory” to proceed. That theory treats the AI model itself as an infringing copy because it embodies transformations of the plaintiffs’ works.¹⁰ The judge said the theory’s viability turns on whether protected works are actually found within the AI systems, a factual question for later stages.¹¹ The court cited Stability AI’s CEO stating that the company compressed 100,000 gigabytes of images into a two-gigabyte file capable of “recreating” them. A summary judgment hearing is scheduled for February 2027.

Disney, Universal, and Warner Bros. v. Midjourney

In June 2025, Disney, Universal, DreamWorks, Marvel, Lucasfilm, and Twentieth Century Fox sued Midjourney in the Central District of California. The 110-page complaint called Midjourney a “copyright free-rider” and showed the platform reliably generating high-quality images of named copyrighted characters from simple prompts.¹² The studios want an injunction that could effectively shut the platform down unless it installs protections against reproducing their intellectual property.

Warner Bros. Discovery filed its own suit in September 2025, alleging Midjourney had made a “calculated and profit-driven decision to offer zero protection for copyright owners.”¹³ The cases were consolidated in November 2025 under the Disney action, with court-ordered mediation to be completed by August 2026.¹⁴

What the AI Cases Could Decide for Everyone Else

Three outcomes matter most for people outside the courtroom. First, whether training a generative model on copyrighted images is fair use or infringement will determine whether companies have to license training data or pay damages for what they already used. Second, whether a trained model itself counts as an infringing copy (the Andersen “model theory”) could expose the models in current circulation, not just the training process. Third, the trademark findings against older Stable Diffusion versions suggest that outputs reproducing watermarks or trade dress carry their own liability, separate from the underlying copyright question.

The UK ruling that model weights are “not themselves an infringing copy” is the clearest early signal, and it favored the AI developer. The US courts have not yet answered the same question, and the Northern District of California has let Getty’s direct infringement claims move toward trial.

Demand Letters for Unlicensed Photos

The lawsuits most people actually encounter have nothing to do with AI. Copyright enforcement firms use image-recognition software to scan the internet for unlicensed photos, then send demand letters on behalf of photographers and stock agencies. The practice is widespread enough to have a nickname: copyright trolling.¹⁵

Firms like Higbee & Associates, PicRights, and CopyCat Legal send thousands of letters a month. A typical letter contains a screenshot of the alleged infringement, a demand for payment in the thousands of dollars, and a short deadline.¹⁶ The number is calibrated to be painful enough to prompt settlement but cheaper than hiring a lawyer and litigating. One New York attorney filed over 1,100 such lawsuits in less than four years.¹⁷

Copyright infringement is a strict-liability offense. Intent does not matter. A business that unknowingly uses an unlicensed photo on its website is just as liable as one that took it on purpose.¹⁵ For registered works, statutory damages run from $750 to $30,000 per infringement, and up to $150,000 if the infringement is willful. Courts can also award attorney’s fees to the prevailing party.¹⁸

A few misconceptions keep feeding these cases. Giving the photographer a credit line is not a substitute for a license. An image without a watermark is still copyrighted. And a business is usually on the hook even when a web designer or marketing firm it hired failed to secure proper rights.¹⁶

If you receive a demand letter, the general guidance is to stop using the image immediately, verify the sender’s authority and whether the work is actually registered, and talk to a lawyer before responding. Settlements for less than the opening demand are common.¹⁵

How Image Copyright Claims Actually Work

Copyright protection attaches automatically the moment a photograph is fixed in tangible form. Under the Berne Convention, which the United States joined in 1989, no registration or copyright notice is required for a work to be protected.¹⁹

Filing a federal infringement lawsuit is different. The 2019 Supreme Court decision in Fourth Estate Public Benefit Corp. v. Wall-Street.com, written by Justice Ginsburg, held that a copyright owner must have a registration certificate from the Copyright Office in hand before suing. Standard processing averages about seven months; expedited processing costs $800 and generally takes five business days.²⁰ The Copyright Claims Board, established by the CASE Act of 2020, is a small-claims alternative for disputes up to $30,000 and accepts a pending application rather than a completed registration.²¹

For online infringement, the DMCA notice-and-takedown system lets copyright owners demand removal without filing suit. A valid takedown notice identifies the copyrighted work, provides the URL of the infringing material, includes a good-faith statement, and is signed under penalty of perjury.²¹ If the alleged infringer files a counter-notice, the platform must restore the content within 10 to 14 business days unless the copyright owner files a federal lawsuit.²²

New Transparency Laws on the Way

Two US bills introduced in early 2026 would give copyright holders more visibility into AI training data.

The TRAIN Act, introduced January 22, 2026, by Representatives Madeleine Dean and Nathaniel Moran, would let copyright holders subpoena AI training data on a certified “good-faith belief” that their work was used, without filing a full lawsuit first.²³
The CLEAR Act, introduced February 10, 2026, by Senators Adam Schiff and John Curtis, would require AI developers to file a detailed summary of every copyrighted work used in training with the Copyright Office at least 30 days before any commercial launch. Violations could draw $5,000 per instance, capped at $2.5 million per action.²⁴

Neither bill had moved past introduction as of mid-2026.

In Europe, Article 53 of the EU AI Act imposes binding transparency rules on providers of general-purpose AI models, including image generators. Providers must publish a detailed summary of their training data using a mandatory European Commission template, disclose licensing arrangements, and explain how they honored opt-outs by copyright holders.²⁵ The AI Office gains enforcement powers on August 2, 2026, with noncompliance penalties up to €15 million or 3 percent of global annual revenue.²⁶

With the first US AI copyright jury trial set for September 2026, the Getty US trial scheduled for January 2028, UK and German appeals in motion, and EU transparency enforcement starting in August 2026, the coming year will produce the first real answers to questions that have been building since generative AI tools reached the public.

1
Houston Law Review. Fair Use and the Origin of AI Training
2
Copyright Alliance. Getty Images v. Stability AI Complaint
3
Manuscript Report. AI Copyright Lawsuits
4
CourtListener. Getty Images (US), Inc. v. Stability AI, Ltd.
5
ChatGPT is Eating the World. Judge Thompson Order Re MTD in Getty Images v. Stability AI
6
Judiciary of England and Wales. Getty Images v Stability AI
7
UK High Court of Justice. Getty Images v Stability AI Judgment
8
Cleary Gottlieb. UK High Court Issues Landmark Ruling in Getty Images v Stability AI
9
Taylor Wessing. Next Steps for Getty v Stability: Why Has Permission to Appeal Been Granted
10
Fstoppers. 5 Legal Battles Will Shape Photography 2026
11
Justia. Andersen et al v. Stability AI Ltd. et al
12
Georgetown Law Tech Institute. Disney, NBC Universal, and DreamWorks File Major IP Lawsuit Against AI Image Generator Midjourney
13
PC Gamer. Midjourney’s Troubles Get Worse as Warner Bros Discovery Sues the AI Image Generator for Copyright Infringement
14
CourtListener. Disney Enterprises Inc v Midjourney Inc
15
Ward and Smith. Crossing the Bridge: Copyright Trolls
16
McNees Wallace & Nurick. Copyright Troll Demand Letters
17
Fennemore Law. The Cost of a Photograph
18
Burke Law. Picture This: A Costly Copyright Infringement Claim Against Your Business for Unauthorized Use of Images
19
Miller Nash. The Accidental Infringer: What You Need to Know About Copyright When Using Photos From the Web
20
Library of Congress Copyright Blog. The Fourth Estate Decision and Copyright Registration
21
U.S. Copyright Office. The Digital Millennium Copyright Act
22
Heimlich Law. DMCA Takedown Notices: Complete Guide for Creators
23
Berkeley Technology Law Journal. The TRAIN Act: Forcing Transparency in AI Training Data
24
Senator Schiff Official Website. Sens. Schiff, Curtis Introduce Bipartisan Bill to Protect Creators’ Work
25
EU Artificial Intelligence Act. Article 53
26
WilmerHale. European Commission Releases Mandatory Template for Public Disclosure of AI Training Data