On November 4, 2025, the UK High Court handed down judgment in Getty Images v. Stability AI,[1] a case emphasized for its significance to content creators and the AI industry and “the balance to be struck between the two warring factions”.[2] Despite significant public interest in the lawsuit, the issues that remained before the court on the “diminished”[3] case were limited (after Getty abandoned its primary infringement claims during trial). The judgment dismisses Getty’s remaining claims of secondary copyright infringement. While some claims of trademark infringement asserted by Getty were upheld, Justice Joanna Smith DBE acknowledged the findings were “extremely limited in scope”.[4]
A Brief Recap of the Case
Getty commenced legal proceedings against Stability AI in 2023. Getty claimed that Stability AI was responsible for infringing a range of its UK intellectual property rights. Getty’s case focused on Stability AI’s Stable Diffusion text-to-image models (including the training and development processes for such models and the outputs generated through their use).
Getty’s primary claims were for copyright infringement and database right infringement under UK law, alleging unauthorised use of its content in the creation of Stable Diffusion. Getty also argued that, in some cases, Getty’s copyright, database right and/or trademarks were infringed by outputs produced by Stable Diffusion users. Getty further alleged that Stable Diffusion was an article which was an infringing copy of Getty’s copyright works (not because the model comprises a reproduction of those copyright works but rather as a consequence of using those copyright works in the training and development process).
Stability AI’s main defence to these claims was that development and training of Stable Diffusion occurred outside the UK, and, with respect to outputs, that there was no copying of substantial parts of underlying images and/or that Getty did not own such works. Getty subsequently abandoned its primary copyright and database right infringement claims after close of evidence at trial. It continued to claim Stability AI was liable for secondary copyright infringement and trade mark infringement (as well as for ‘passing off’).
Secondary Copyright Infringement Claim Dismissed
Getty claimed that Stability AI was liable for secondary copyright infringement under sections 22 and 23 of the UK Copyright, Designs and Patent Act 1988 (“CDPA”),[5] by virtue of the importation of Stable Diffusion (through its downloading in the UK) and its distribution in the course of business via the Hugging Face platform.
Justice Joanna Smith DBE noted that the “battle lines” for this claim “can be readily and simply identified”.[6] Such claim required that the court find that Stable Diffusion is (1) an “article” under sections 22 and 23 CDPA, and (2) an “infringing copy” for the purposes of section 27(3) of the CDPA (i.e., because Stable Diffusion had been imported into the UK and its making in the UK would have involved copyright infringement of Getty’s copyright work).
The court concluded that the concept of an “article” should be understood to cover more than just tangible goods and could include electronic copies stored in an intangible form (siding with Getty on this matter). However, Getty’s claim was premised on the fact that Stable Diffusion was an “infringing copy” merely because its making (i.e., repeated exposure of the model weights of Stable Diffusion to training data comprising Getty’s copyright works) would have constituted infringement if done in the UK. The court rejected this argument and held that for an article to be an “infringing copy” it must have at some point consisted of/stored/contained a copy of a copyrighted work. As the court found that this was not the case for Stable Diffusion, it dismissed Getty’s secondary copyright infringement claim.
Limited Trade Mark Infringement Found
Getty also pursued a claim for infringement under sections 10(1)–(3) of the UK Trade Marks Act 1994 (“TMA”)[7] on the basis that Stable Diffusion outputs contained versions of Getty and iStock watermarks. These claims were supported by example outputs generated “‘in the wild’ by third parties” and through sample testing conducted by Getty in the course of the litigation.[8]
The judge considered the trade mark claims in the context of different versions of Stable Diffusion (including those pleaded as having been used to generate the example outputs) and examined the evidence for watermark generation in each case. The judge held that, on balance, at least one UK user would have generated a synthetic image containing a Getty or iStock watermark when using older versions of Stable Diffusion (i.e., v1 and v2). However, the judge dismissed all trade mark claims in respect of Stable Diffusion XL and v1.6 having regard to the fact that (i) there was “contemporaneous evidence of steps being taken to address the watermark issue prior to release” of the models[9] and (ii) no evidence of the generation of watermarked images ‘in the wild’ was advanced by Getty for these models (which the judge noted to be “strongly indicative of the fact that… this issue was resolved using appropriate filters applied to the training data”).[10]
The judge then compared the watermark examples produced in evidence against the Getty and iStock watermarks, identifying just a few specific examples which constituted trade mark infringement under Sections 10(1) or 10(2) of the TMA. The judgement emphasised that it was “impossible to know how many (or even on what scale) watermarks have been generated in real life that would fall into a similar category”[11] and, accordingly, the trademark infringement findings are “extremely limited in scope”.[12]
Getty’s claim under Section 10(3) of the TMA was dismissed entirely on the basis that no evidence was produced to demonstrate that use of the Getty and iStock trademarks gave rise to relevant injury (i.e., detriment to the distinctive character or repute of the trademark, or unfair advantage being taken of such trademark).
Key Takeaways
- The findings in this case were highly fact-specific, and caution is needed when extrapolating principles from the ruling. The analysis in the judgment focuses on the specific technical architecture, capabilities and output of the Stable Diffusion models. It will likely be difficult to analogize the exact circumstances of this case, analyzed in depth in the judgement, to other AI models and the outputs generated through their use.
- The judicial finding that Stable Diffusion did not contain, store, or consist of, a copy of any copyrighted work could have important implications for interpretation and application of regulation in an AI context. In this case, the fact that the development process for Stable Diffusion involved the reproduction of copyright works (through storing images locally and in cloud computing resources and then exposing the model weights to those images) was “of no relevance”.[13] This may weigh against a regulatory view that models ‘contain’[14] the data on which they are trained.
- In the alternative, the judge also set out certain findings of fact premised on the assumption that Stable Diffusion is an “infringing copy.” If, contrary to the findings in this judgment, Stable Diffusion is capable of being an infringing copy, the judge concluded that distribution of Stable Diffusion by making it available for download in the UK amounts to an act of secondary infringement. This issue will be one to watch for on any appeal.
- The judgment brings into focus the importance of location of training. The fact that copyright law is not extra-territorial presented a clear barrier in this case to bringing primary copyright infringement claims in the UK. Navigating this challenge has been a key priority for regulators and was discussed in both the UK Intellectual Property Office’s December, 2024 Copyright and AI Consultation[15] and the European Parliament’s July, 2025 study on generative AI and copyright law.[16]
- Trade mark infringement at the output stage is a potential hook in specific cases, but this will likely depend on the sophistication of the model in question (and relevant technical guardrails). The availability of evidence to support a claim of this type will also be key – and it will likely be difficult to contend that infringement of this kind is widespread.
- While Getty’s case narrowed during trial, Getty is still pursuing primary copyright infringement claims in its litigation against Stability AI in the U.S. Other significant U.S. AI-copyright infringement litigation includes Andersen v. Stability AI Ltd. et al, 3:23-cv-00201 (N.D. Cal. Oct. 30, 2023). Note, Cleary Gottlieb represents Midjourney in this matter.
[1] Getty Images v Stability [2025] EWHC 2863 (Ch).
[2] Ibid. [13].
[3] Ibid. [13].
[4] Ibid. [757].
[5] The text of the CDPA can be found here.
[6] Getty Images v Stability [2025] EWHC 2863 (Ch) [548].
[7] The text of the TMA can be found here.
[8] Getty Images v Stability [2025] EWHC 2863 (Ch) [67].
[9] Ibid. [245].
[10] Ibid. [243].
[11] Ibid. [758 (iv) and (v)].
[12] Ibid. [757].
[13] Ibid. [600].
[14] See, for example, European Data Protection Board Opinion 28/2024 on certain data protection aspects related to the processing of personal data in the context of AI models (adopted on 17 December 2024), [38].