The High Court in London has narrowed one of the most closely watched legal fights around generative AI. Getty Images' main copyright case against Stability AI has been thrown out, leaving a ruling that matters for how courts may treat AI models built from large image collections.
The decision centered on Stable Diffusion and whether the model itself could be treated as an unlawful copy because of how it was created. The court's answer was no, based on the facts before it.
The case Getty Images brought
Getty Images accused Stability AI of "scraping" millions of its photos to train Stable Diffusion. Getty Images described the system as an "existential threat" to the creative industry, framing the dispute around both copyright and the commercial impact of AI-generated images.
But the shape of the case changed before the court reached its decision. Getty Images dropped its main claims, including claims about the training of the model and the images Stable Diffusion produces.
That mattered because the court documents said there was no evidence that the model training happened in the UK. Once that point was removed, the remaining dispute became narrower. The court was left to assess questions of secondary copyright and trademark infringement.
In practical terms, the ruling did not decide every possible question about AI training and copyright. It addressed the arguments still before the High Court in London after Getty Images' broader claims had fallen away.
Why the copyright claim failed
The central copyright question was whether Stable Diffusion itself counted as an "infringing copy." Getty Images argued that it did because "the making of its model weights would have constituted infringement of the Copyright Works had it been carried out in the UK."
Judge Joanna Smith rejected that reasoning. Her ruling said that an AI model such as Stable Diffusion, which "does not store or reproduce any Copyright Works (and has never done so)," is not an "infringing copy" under the UK Copyright, Designs and Patents Act (CDPA).
The distinction is important. Getty Images' argument depended on treating the model as the relevant copy. The court instead focused on whether the model stored or reproduced the copyright works at issue. On the court's view, the answer was that it did not.
The ruling also addressed the question of whether the law could cover intangible objects as "articles." Even with that possibility, Getty Images' argument did not succeed. The court did not accept that Stable Diffusion itself fell within the copyright claim in the way Getty Images needed.
A narrower trademark finding remained
Getty Images did not leave the case with nothing. The court found a limited trademark issue involving some older versions of Stable Diffusion. In certain cases, those versions could create watermarks similar to Getty Images' or iStock's trademarks.
That finding was specific. It was tied to particular image examples, not a broad conclusion about every use of Stable Diffusion or every version of the system.
The judge also emphasized the limits of the evidence. The ruling said "it is impossible to know how many (or even on what scale) watermarks* have been generated in real life that would fall into a similar category" matching the pattern before the court.
That left Getty Images with a narrow win, but not the wider outcome it had sought. The court dismissed Getty Images' claims for reputational harm and said there was no basis for additional damages.
What the ruling means for AI companies
For Stability AI, the headline outcome is clear: Getty Images' main copyright case did not survive. The court did not accept that Stable Diffusion was an "infringing copy" merely because of Getty Images' theory about how the model weights were made.
For generative AI companies, the ruling is a significant moment because it separates the model itself from the copyrighted images Getty Images said were used in training. The court's reasoning turned on whether the model stored or reproduced the copyright works, and the ruling said it did not.
At the same time, the decision does not erase every legal risk around AI image generators. The trademark finding shows that outputs from older versions can still matter when they resemble protected marks in specific examples.
The remaining lessons are therefore mixed:
- Copyright: The court rejected the argument that Stable Diffusion itself was an "infringing copy" under the CDPA.
- Training evidence: The case was narrowed because there was no evidence that model training happened in the UK.
- Trademark: Some older versions of Stable Diffusion could generate watermarks similar to Getty Images' or iStock's trademarks in certain cases.
- Damages: The court dismissed reputational harm claims and found no basis for additional damages.
The legal fight narrowed, but did not disappear
The High Court in London did not issue a sweeping answer to every copyright question raised by generative AI. It ruled on the case as it stood after Getty Images dropped major claims and after the evidence narrowed the dispute.
Still, the outcome gives AI developers an important result: a model that "does not store or reproduce any Copyright Works (and has never done so)" was not treated as an "infringing copy" on the facts before the court.
For rights holders, the case also shows how much depends on evidence, jurisdiction, and the precise claim being made. Getty Images' broad copyright challenge fell away, while a narrower trademark issue survived only in limited form.
The result is not a full victory for either side on every issue. It is a ruling that trims the case down to what the court could prove and decide: Stable Diffusion was not an "infringing copy," but some older outputs raised a limited trademark problem.