Getty Images v Stability AI: High Court Judgment

Since 2023, the legal battle between various Getty Images entities (“Getty“) and Stability AI (“Stability“) has drawn the attention of those interested in the consistency of AI models being trained on third party content (and the related intellectual property rights), culminating recently in a highly anticipated judgment from the High Court providing judicial insight into the application of UK intellectual property laws in respect of generative AI models.

Background

In 2023, Getty brought various claims against Stability, covering primary copyright infringement, database right infringement, trade mark infringement and passing off, alleging that Stability had infringed Getty rights by using millions of Getty’s images, without Getty’s permission, to train Stability’s text-to-image AI model, ‘Stable Diffusion’.  Such use, Getty argued, also meant that Stable Diffusion itself was an article (i.e. not a directly infringing copy) the use of which constituted secondary infringement of Getty’s rights.  Further, Getty alleged that some outputs of Stable Diffusion resembled Getty’s protected content (with some images even displaying Getty watermarks).

In the course of the proceedings, Getty, having struggled to produce sufficient evidence that the training (and therefore the copying) took place within the UK, dropped many of its claims (including for primary infringement) and the claims were narrowed down to:

1. secondary copyright infringement – by making Stable Diffusion available for download in the UK, Stability had imported, possessed and dealt with an “article” which was (and which Stability knew or ought to have known was) an “infringing copy” of its protected works (as, had it been made in the UK, its making would have constituted copyright infringement); and
2. trade mark infringement – that certain outputs of Stable Diffusion infringed Getty’s trade marks (due to displaying features that looked identical or similar to Getty’s watermarks).

The decision

In her decision of 4 November 2025, Justice Joanna Smith DBE ruled mostly in favour of Stability, largely dismissing Getty’s claims with Getty only having limited success in respect of its trade mark claim.  Justice Smith found that:

1. Stability AI did not commit secondary copyright infringement – although there was evidence that Getty’s images were used to train Stable Diffusion (albeit outside of the UK), the trained AI model did not store or reproduce (i.e. copy) Getty’s copyright works in itself and therefore did not itself constitute an article infringing copyright.
2. Stability committed limited infringement of Getty’s trade marks – a small number of outputs produced by early versions of Stable Diffusion were found to have features which displayed or resembled Getty’s watermarks.

It is possible that Getty will seek to appeal this judgment (with an appeal deadline in the coming months), however, given Getty dropped various aspects of their claim at the tail end of the hearing Getty’s options in the UK may be limited.

Getty continues to pursue a parallel matter against Stability in the US which may be a more appropriate jurisdiction for Getty’s key claims including direct infringement (given the direct copying occurred in the US).

Stay up to date with:

• Trending Topics
• Latest Insights
• Upcoming Events
• Company Updates

Follow us on LinkedIn

Take Me There

Observations

While Getty’s limited success may be seen as a blow to those whose intellectual property rights have been used to train AI models, the judgment was interesting for many reasons, and we would note the following.

1. The key question is still unclear – as the training of the AI model did not take place in the UK, due to issues around jurisdiction and evidence, the court was unable to confront questions around the legality of mass scraping of copyright works to train AI models in the UK. This case does not provide a precedent for all forms of AI models either as cases involving AI models which store or reproduce copyrighted works may be treated differently. Developers of AI should therefore remain cautious and pay attention to future litigation and guidance, taking legal advice and support where appropriate.
2. “Article” is broad enough to include intangible information – “article” does not have an express definition under relevant legislation and, historically, has been understood to only extend to tangible objects. Justice Smith’s taking into account changes to circumstances since legislation was enacted to conclude that “article” extends to intangible information, produced a clarification that, for the purpose of copyright, intangible articles such as AI models, are capable of being found to be infringing, in the same way as tangible articles.
3. Trade mark specifications – synthetically generated AI images are not ‘photographs’ –  Getty had sought to argue that outputs from Stable Diffusion could be characterised as “photographs” as that is what users would typically call such outputs (and would have given Getty greater scope to claim trade mark infringement). Justice Smith’s view, that outputs were synthetically generated images, serves as a reminder that, when preparing trade mark specifications, the relevant function and output should be considered very carefully (in a literal sense) and appropriate scope used to ensure claims are not narrowed down (particularly for businesses with novel technologies).

Conclusions

As the judgment rules that AI models which do not store or reproduce copyright works will not be found to be infringing articles under the UK’s copyright regime, AI developers may take reassurance from this judgment with creative rights holders still in an unsatisfactory position (given their view that the training of AI models involves en masse copyright infringement) in the UK.

This case does not provide a sweeping ‘green light’ that all training of AI on copyright works will be lawful in the UK, non-infringement will ultimately depend on the architecture of each model, how it is developed and used as well as the findings of future cases in the UK and overseas.

Ultimately, the judgment serves as a reminder of the growing tensions between innovation in AI, and the UK’s increasingly outdated legislative intellectual property frameworks, and is the most recent step in the long road to clarity on matters concerning intellectual property in generative AI.  A ‘process update’ from the government’s expert working groups on copyright and AI is expected before the end of the year which may shed light on how this issue may develop further.

Many developers and rights holders may still be uncertain of their position in light of this judgment – our digital and technology experts and intellectual property lawyers can help you gain clarity of your situation.

This article is co-authored by James Talbot and Mridvika Elliott.