200-page High Court judgments on highly technical matters of copyright and trade mark law rarely make national headlines. Mrs Justice Joanna Smith’s judgment in Getty Images v Stability AI [2025] EWHC 2863 (Ch), however, was an exception.

It is a judgment that was long awaited, having been delivered some two and a half years after Particulars of Claim were served. It was also hotly anticipated, as the case represented the first of its kind to address the question – often asked but still lacking a clear answer – of how various intellectual property laws apply to the training and deployment of AI models.

However, that anticipation dampened as the litigation proceeded and it became clear that what was one of the key questions – namely whether the training of AI models on works protected by copyright constitutes infringement in the UK – would not be answered because the training took place outside the jurisdiction. This left a residual case of secondary copyright infringement that turned on nuanced, technical matters, and claims for trade mark infringement, findings about which the judge herself described as “both historic and extremely limited in scope”.

The result is a judgment that – whilst offering valuable clarity about how elements of existing IP law apply to AI models – will, for some, feel as though it has not lived up to the headlines. The central question about whether the training of AI models on copyright works (and their outputs) constitutes copyright infringement under UK law remains open.

The case revolved around the AI image generation tool, Stable Diffusion, which was developed and distributed by Stability AI. In early 2023, Getty Images (Getty) issued a claim against Stability AI arguing that the training of Stability Diffusion involved the unauthorised scraping of millions of works within Getty’s database of which it was either the copyright holder or exclusive licensee.

Initially, Getty’s case included claims of primary copyright infringement, both in respect of Stability AI’s use of its works to train its model, and the outputs it produced. However, those arguments were subsequently abandoned by Getty because, as the judgment sets out, “(i) there is no evidence that the training and development of Stable Diffusion took place in the United Kingdom… (ii) the prompts which it was alleged had been used to generate the examples of infringing output from the Model in evidence in these proceedings have been blocked by Stability such that the relief to which Getty Images would have been entitled in respect of their allegations of primary infringement of copyright has now been substantially achieved.” Getty’s database rights claim, which was inherently linked to its primary copyright infringement claims, also fell away.

That left three remaining arguments: (1) secondary copyright infringement under sections 22 and 23 of the Copyright, Designs and Patents Act 1998 (CDPA); (2) trade mark infringement under sections 10(1), 10(2), and 10(3) of the Trade Marks Act 1994 and (3) passing off.

The judgment is considerable in both its length and discussion of the technical aspects of how Stable Diffusion works. We cannot hope to address all the points that the judge addressed in this post but we concentrate instead on her principal findings in relation to Getty’s claims.

Having dropped its claims for primary copyright infringement, Getty fell back on an argument positing that Stability AI had committed two separate but related acts of secondary infringement of copyright.

In particular, Getty relied on sections 22 and 23 CDPA which provide that copyright in a work is infringed by a person who, without licence of the copyright owner, imports, possesses, sells or distributes an ‘article’ which he knows or has reason to believe is an ‘infringing copy’ of the work.

It was Getty’s position that the Stable Diffusion model itself – or more accurately its model weights – constituted an ‘article’ for the purposes of these provisions. Furthermore, whilst having accepted that the model weights did not themselves store a copy of a particular work, Getty nonetheless relied on s.27 CDPA which provides that an article is an infringing copy if “its making in the United Kingdom would have constituted an infringement of the copyright in the work in question, or a breach of an exclusive licence agreement relating to that work”.

In other words, given that it was commonly accepted that the model trained on Getty’s copyright-protected works (albeit not in the United Kingdom) and that that training involved the reproduction (by way of storage) of those works, the ‘making’ of the model weights would have been an infringement for the purposes of section 27 CDPA if it had taken place in the United Kingdom.

To this, Stability argued that (1) the definition of an ‘article’ for the purposes of the relevant provisions was limited to tangible objects, and (2) an article cannot be an ‘infringing copy’ if  it does not contain a copy of the relevant copyright work.

The judgment explores these arguments in significant detail, recognising that they raised “two novel issues of law”.

On the question of whether an ‘article’ must be in tangible form, the judge ultimately sided with Getty’s view. As she explained, “I consider that an article, which must be an infringing copy, is capable of being an electronic copy stored in intangible form. Standing back, I agree with Getty Images that if the word “article” were construed as only covering tangible articles, this would deprive authors of protection in circumstances where the copy is itself electronic and it is then dealt with electronically. Not only would that be inconsistent with the words of the statute, but it would also be inconsistent with the general scheme of copyright protection which is to reward authors for their creative efforts”.

However, for Getty to be successful, the judge would also have to agree with its position that the model weights (as an ‘article’) also constituted an ‘infringing copy’. On this point, Stability AI argued that an infringing copy must necessarily contain a copy of the relevant copyright works. Getty, on the other hand, argued that the legislation merely required that its making constituted an infringement. As the judge explained, Getty’s submission was that there is “no requirement that the article thus made must continue to retain a copy or copies of the work”.

Mrs Justice Smith accepted Stability’s position, explaining her reasoning as follows:

“In reality therefore, the dispute between the parties as it finally emerged in closing, really turns on whether an article whose making involves the use of infringing copies, but which never contains or stores those copies, is itself an infringing copy such that its making in the UK would have constituted an infringement. Taking the specific facts with which I am concerned, is an AI model which derives or results from a training process involving the exposure of model weights to infringing copies itself an infringing copy?

In my judgment, it is not. It is not enough, as it seems to me, that (in Getty Images’ words) “the time of making of the copies of the Copyright Works coincides with the making of the Model” (emphasis added). While it is true that the model weights are altered during training by exposure to Copyright Works, by the end of that process the Model itself does not store any of those Copyright Works; the model weights are not themselves an infringing copy and they do not store an infringing copy. They are purely the product of the patterns and features which they have learnt over time during the training process. Getty Images’ central submission that “as soon as it is made, the AI model is an infringing copy” is, accordingly, in my judgment, entirely misconceived…

I agree with Stability that the concept of an infringing copy cannot be interpreted in the abstract without reference to the fundamental nature of a copy, as revealed by section 17. Such an interpretation precludes the potential for an article which has never contained a copy to be capable of being “an infringing copy”. But, sections 27(2) and (3) require the article that is made to be a “copy”. They are not concerned with a process which (while it may involve acts of infringement) ultimately produces an article which is not itself an infringing copy”.

It followed, therefore, that Getty’s claims of secondary infringement failed.

Getty was more successful in its claims for trade mark infringement, although even that success was somewhat limited.

As many will be aware, Getty employs watermarks in its library, affixing the words “getty images” or “iStock” to the works. These are registered trade marks, which Getty argued had been reproduced in outputs of Stable Diffusion. It therefore brought claims under sections 10(1), 10(2) and 10(3) of the Trade Marks Act 1994 (TMA), as well as for passing off.

Despite the fact that much of the commentary about this case concerned the findings in relation to copyright infringement, the majority of the judgment deals with the claims of trade mark infringement, and it does so in considerable detail. In part this is because much time is dedicated to determining which versions of the Stable Diffusion model generated outputs with watermarks that equated to the trade marks, how legible the reproductions had to be to amount to infringement, and whether, in fact, the watermarks were likely to have been generated in real life by users in the UK.

The judge also had to contend with Stability AI’s argument that, to the extent the watermarks were “used” for the purposes of the TMA, the use was by the end users of the model, rather than Stability AI itself. The court rejected this argument, finding that “the only entity with any control in any meaningful sense of the word over the generation of watermarks on synthetic images is Stability. It is certainly not “passive” as Stability submits”.

Ultimately, the judge found that there was infringement under s.10(1) TMA in respect of the ‘iStock’ watermarks, but only in relation to earlier versions of Stable Diffusion. No infringement under s.10(1) was found to have taken place as regards the ‘getty images’ watermark either because there was no evidence that it has been reproduced ‘in the real world’ in any Stable Diffusion versions or, where there was such evidence, the watermark was not identical to the ‘getty images’ marks.

As for s.10(2) TMA, the judge found that there was infringement in relation to both the ‘iStock’ and ‘getty images’ marks, but again only in relation to earlier versions of the Stable Diffusion model. She explained that, depending on how the average consumer accessed the Stable Diffusion model, they might either consider that the synthetic outputs generated by Stable Diffusion which contained the watermarks were supplied by Getty, or would assume that they were a result of “some form of licensing or other economic agreement in place with Getty Images”. However, the judge was quick to point out that that her findings in relation to s.10(2) infringement were “highly fact sensitive”, applied to three specific examples that she was presented with, and should not be generalised.

Finally, the judge held that there was no infringement under s.10(3) TMA. Much of this was due to the fact that, as the judge put it, she had “no means of determining the scale of the generation of infringing watermarks [by earlier versions of the model] and not a scrap of evidence of any change in economic behaviour by the users of those Models”. This resulted in conclusions that there was no basis on which she could infer that Getty’s trade marks would be diluted, or that the reputation of the marks would be harmed. Finally, there was no evidence that Stability AI intended the watermarks to appear in its outputs (the judge found that the opposite was true) or wished to take advantage of Getty’s reputation, so the claim for unfair advantage also failed.

As for passing off, in the light of her findings on trade mark infringement, she took up the invitations of the parties not to address the point further.

As a result, Getty’s claims for trade mark infringement were successful in part. However, the judge neatly summarises her findings as “both historic and extremely limited in scope”.

To read the judgment in full, click here.