Getty Images v Stability AI: UK High Court Finds No Secondary Copyright Infringement in Landmark GenAI Ruling

In its highly-anticipated judgment in Getty Images v Stability AI [2025] EWHC 2863 (Ch), the UK High Court rejected Getty Images’ claim for secondary copyright infringement against Stability AI.

In short, the issue at trial was whether the importation and distribution of Stability AI’s “Stable Diffusion” image-generation AI model in the UK amounted to secondary infringement of Getty Images’ copyright works (namely, high-quality photographic images), on the basis that (Getty Images claimed) the making of the Stable Diffusion model weights would have constituted infringement of those copyright works had that process been performed in the UK.

On the evidence, the Stable Diffusion model weights never contained or stored any infringing copies of Getty Images’ copyright works. The Court held that an infringing copy (for the purposes of secondary infringement in the UK) must be a copy, in that it must have consisted of or stored or contained a copy of a copyright work. For that reason, the secondary copyright infringement claim failed.

Getty Images’ claim for trademark infringement, made on the basis that at least some images generated by Stable Diffusion contained the Getty Images watermark, was partially successful.

This is one of the first judgments to be handed down among the many copyright infringement claims which are currently pending in the generative AI (“GenAI”) space around the world, and has garnered significant international attention. However, the case narrowed significantly before trial, meaning that two important issues (copyright infringement during the training and development of AI models and copyright infringement by AI-generated outputs) remain unconsidered by the UK courts.

The secondary copyright infringement claim

Sections 22 and 23 of the Copyright, Designs and Patents Act 1988 (“CDPA”) provide that certain acts committed in the UK (including importation and distribution) will amount to copyright infringement if performed in relation to an “article” which is (and which that person knows or has reason to believe is) an “infringing copy” of a copyright work.

Section 27(3) of the CDPA contains the following definition:

“An article is also an infringing copy if—

(a) it has been or is proposed to be imported into the United Kingdom, and

(b) its making in the United Kingdom would have constituted an infringement of the copyright in the work in question […].”

Getty Images’ claim at trial was not that the Stable Diffusion model itself was a copy of, or stored any copies of, Getty Images’ copyright works. Instead, the claim was that Stable Diffusion’s model weights were “articles” (albeit intangible ones) whose creation involved copyright infringement.

Getty Images’ claim therefore raised two novel questions of law:

(i) whether the model weights were “articles” within the meaning of Sections 22 and 23 of the CDPA; and

(ii) whether they were “infringing copies” within the meaning of Section 27 of the CDPA.

The Court found that “articles” included intangible articles, such that the model weights were “articles”. However, the Court held that the model weights were not “infringing copies”.

It was common ground that the training of the Stable Diffusion image-generation AI model, which took place in the US, involved the reproduction (by storage) of Getty Images’ copyright works. It was also common ground that the making of the model weights required repeated exposure to the training data which contained the copyright works. However, the evidence was that the Stable Diffusion model weights did not themselves constitute or contain copies of training data (including any of Getty Images’ copyright works).

Getty Images argued that this did not matter, because the making of the model weights would have constituted infringement had it been carried out in the UK.

The judge disagreed, finding that an “infringing copy” must be a copy, and observing that “I cannot see how an article can be an infringing copy if it has never consisted of/stored/contained a copy”.

It was not enough that the model weights were altered during training by exposure to copyright works, because at the end of that process, the model weights were not infringing copies and did not store any infringing copies.

Recent developments

Whilst the significance of this case somewhat diminished as a result of the issues having been narrowed before trial, the decision nonetheless illustrates the importance to rights holders of understanding precisely how a given generative AI model utilises their copyright works. Of course, the majority of pending GenAI cases relate to the use of copyright works in training data, and it is hoped that questions relating to the use of Getty Images’ copyright works in training Stable Diffusion will be resolved in the ongoing US proceedings between the two companies.

Notably, the Getty Images v Stability AI decision demonstrates the potential for divergence between jurisdictions: whilst it is not binding on the UK Courts, the recent landmark decision of the 42nd Civil Chambers of the Munich Regional Court in GEMA v OpenAI (Case No. 42 O 14139/24) found that OpenAI’s use of song lyrics did amount to copyright infringement. Whether a harmonised approach to the use of copyright works by generative AI models can be achieved will therefore continue to present a challenge to the international legal community.

Getty Images has been granted permission to appeal to the UK Court of Appeal on the question of secondary copyright infringement.