getty images v stability ai: Getty v. Stability AI

For two years, one lawsuit carried more weight than almost any other in the creator world. Getty Images — the giant stock-photo library — versus Stability AI, the company behind the image generator Stable Diffusion. It was supposed to be the case that answered the question every artist, photographer, and designer has been asking: can an AI company hoover up your work to train its models without asking?

Then, in November 2025, the UK court gave its answer. And the headlines were brutal for Getty: it lost. But if you stop at the headline, you’ll draw exactly the wrong conclusion. Here’s what actually happened, in plain English — and what it really means for you.

A gold-framed photo dissolving into particles that reform as a neural network, symbolising images used to train AI — At the heart of the case: millions of images allegedly pulled in to train an AI model.

The 30-second version

Getty sued Stability AI in two places at once: the United Kingdom and the United States. The claim, in essence, was that Stability copied around 12 million Getty images — without a licence, without payment — to train Stable Diffusion, and that the model then competed directly with the very library it had learned from. As proof that Getty’s photos were in the training data, Getty pointed to something startling: some AI-generated images came out with a mangled version of the Getty Images watermark still stamped across them.

Track the cases yourself

Every current AI-copyright ruling — and the ones still being fought — lives in our AI Copyright Ruling Tracker. Filter by your creator type to see, in plain language, how each case affects your rights.

In the UK, the case went all the way to trial in mid-2025. The judgment landed on 4 November 2025. Getty lost almost everything it had left — but won one narrow, interesting point. In the US, the fight is still very much alive. Neither case has actually decided the big question. That nuance is the whole story.

What Getty accused Stability AI of doing

Strip away the legal language and Getty made three basic accusations:

Copyright infringement in training — copying millions of Getty’s photographs to build the model in the first place.
Copyright infringement in the outputs — the model spitting out images that reproduced Getty’s protected works.
Trademark infringement — generated images carrying a distorted Getty or iStock watermark, which could fool people into thinking Getty was the source or had approved the image.

On paper, that’s a powerful combination. The problem for Getty wasn’t that these claims were unreasonable — it was where it tried to prove them.

Why Getty mostly lost the UK case

Here’s the twist most headlines skipped: Getty abandoned its main copyright claims partway through the UK trial. Not because it changed its mind about being wronged, but because copyright law is territorial — UK law only governs things that happen in the UK. And Getty couldn’t show that the actual copying and training had taken place on British soil. Stability had trained its models on cloud infrastructure in the United States. No UK copying, no UK copyright claim.

A gold balance scale weighing photographs against an AI model above a dashed border line, symbolising a jurisdiction-driven ruling — The UK outcome turned less on right-and-wrong than on where the training actually happened.

That left two narrower issues for the judge, Mrs Justice Joanna Smith, to decide.

The first was “secondary” copyright infringement — the idea that importing the trained model into the UK was like importing an illegal copy of a copyrighted work. The court said no. A trained AI model is a set of mathematical weights; it does not store or contain the original images. Because the model isn’t itself a copy of Getty’s photographs, bringing it into the UK, possessing it, or distributing it there wasn’t secondary infringement. That’s a genuinely significant technical finding for how the law views AI models.

The second issue was trademark — and this is where Getty scored its one win. The court accepted that some Stable Diffusion outputs reproduced the Getty and iStock watermarks, and that in limited situations this could confuse customers about who made or endorsed the image. That’s trademark infringement. Crucially, the judge held that Stability AI — not the user typing the prompt — was responsible, because Stability controlled the data the model was trained on. Getty later described this as confirming that liability for a watermark appearing in an output “lies with the model provider.”

So the scoreboard: copyright claims dropped or rejected, one limited trademark win, and Getty granted permission to appeal in December 2025.

The most important thing the ruling did NOT decide

This is the part to tattoo on your brain before you repeat anything you read about this case: the UK judgment did not decide whether training an AI on copyrighted work is legal.

Because Getty dropped the training claims over the jurisdiction problem, the judge never ruled on the central question. The outcome wasn’t “AI training is fine.” It was closer to “Getty sued in the wrong country to test that question.” Anyone telling you this case gave AI companies a green light to scrape your portfolio is reading a verdict that was never handed down.

The US case is a different animal

Getty’s American lawsuit started in Delaware back in February 2023. In August 2025, Getty voluntarily dismissed the Delaware case and re-filed the same day in the Northern District of California — home turf for the tech industry and its courts. That case is ongoing.

Getty’s one win: its watermark showing up inside generated images was treated as a trademark problem.

Why does the US case matter more for the big question? Two reasons. First, the jurisdiction problem largely goes away — the training and the company’s US operations are squarely within reach of US law. Second, the US has a doctrine the UK doesn’t: fair use. The knock-down fight over whether training a generative model on copyrighted images is “transformative” fair use — or plain infringement — is one American courts will actually have to answer. Early rulings have already let Getty’s trademark claim move forward, finding it had adequately argued both commercial use and a likelihood of confusion. The copyright core is still to come.

In other words: the case that could set the real precedent is the one still in progress, not the one that just ended.

What this actually means for you

You’re not Getty — you don’t have a legal war chest. But this case hands working creators several practical lessons:

Don’t believe the “AI won” headline. The UK loss was about jurisdiction and a technical point on how models store data, not a ruling that scraping your work is lawful. The core question is unresolved.
Watermarks and brands are a real weapon. The one claim Getty won was the trademark one. If an AI tool generates something carrying your logo, brand name, or watermark in a way that could confuse people, that’s a trademark issue — and the model provider, not just the user, can be on the hook.
Where things happen matters. Copyright is territorial. As a creator you can’t control where a company trains its model, but it shapes which laws even apply — and why the US cases carry the most weight.
The precedent that will affect you most hasn’t been set yet. Keep your eye on the US fair-use rulings, not just the UK result.

A gold shield with a seal in front of a stack of creative works, symbolising protecting your work — The practical lesson for creators: watermarks, trademarks, and records are still your strongest tools.

What to do now

You can’t litigate like a stock-photo giant, but you can borrow the parts of its strategy that worked:

Watermark and sign your published work. Beyond deterrence, a visible mark is both evidence your work was taken and a potential trademark hook if it resurfaces in AI outputs.
Register the trademarks that matter. Your business name, logo, and brand marks are worth registering — trademark was the one theory that landed a hit here.
Keep provenance records. Save originals, timestamps, and posting history. If you ever need to prove a work is yours and was taken, that record is your case.
Watch the US dockets, not the hot takes. The rulings that will genuinely shape creators’ rights are still being written in American courtrooms.

Getty v. Stability AI was supposed to be the verdict on AI and creative work. It turned out to be something more modest and more honest: a reminder that the biggest questions are still open, that watermarks and trademarks have real teeth, and that the fight for creators’ rights is a marathon, not a single headline.

This article is general information, not legal advice. AI and copyright law is moving fast and differs by country and state — for a specific situation, check the current law where you are or talk to a qualified attorney.

IPSurge helps creators and small businesses understand and protect their intellectual property — the ideas, names, and work that make them unique.

Sources & further reading:

Getty v. Stability AI Explained — What It Means for Creators