Getty Images v Stability AI: Trade mark claims succeed whilst AI copyright case dismissed

In Getty Images (US) Inc & Ors v Stability AI Limited [2025] EWHC 2863 (Ch), Mrs Justice Joanna Smith DBE delivered a significant judgement addressing intellectual property claims against generative artificial intelligence. The case concerned Stability AI's Stable Diffusion model, a latent diffusion AI system that synthesises images from text prompts or seed images by modelling probability distributions based on training data.

Getty Images brought claims for primary and secondary copyright infringement, database right infringement, trade mark infringement and passing off against Stability AI. However, shortly before closing submissions, the claimants abandoned various aspects of their claim, substantially narrowing the issues for determination. The judgement nevertheless addresses matters of importance for intellectual property law in the context of AI model deployment.

The court found that Stability bears no direct liability for tortious acts arising from the release of version 1.x Models via CompVis GitHub and CompVis Hugging Face pages. This finding has implications for understanding legal responsibility in open-source AI development.

Trade mark infringement succeeded only in limited circumstances. Getty Images established section 10(1) infringement of the ISTOCK Marks in respect of watermarks generated by users of v1.x Models accessed via DreamStudio or the Developer Platform, based specifically on watermarks shown on the Dreaming Image and Spaceships Image. Under section 10(2), infringement of the ISTOCK Marks was similarly established for v1.x Models, whilst infringement of the Getty Images Marks was found in respect of watermarks generated by v2.x users, specifically regarding the First Japanese Temple Garden Image generated by Model v2.1.

However, Getty Images failed to establish section 10(1) infringement concerning Getty Images watermarks. The section 10(3) claim was dismissed entirely. Crucially, no evidence demonstrated any UK user generating either Getty Images or iStock watermarks using SD XL and v1.6 Models, resulting in dismissal of claims relating to those models.

The court declined to address passing off allegations, given the outcomes on other causes of action.

The secondary copyright infringement claim was dismissed, representing a significant determination for AI development. Mrs Justice Smith concluded that whilst an 'article' may constitute an intangible object under the Copyright, Designs and Patents Act 1988, an AI model such as Stable Diffusion, which neither stores nor reproduces Copyright Works, cannot be an 'infringing copy'. Consequently, no infringement arises under sections 22 and 23 CDPA. This analysis holds particular significance given the jurisdictional limitations identified in the judgement. The court noted that Getty Images' case regarding model training might be maintainable in the jurisdiction where such training occurred, but no basis existed for such claims in England and Wales under the 2006 Regulations.

On ancillary matters, Getty Images failed to establish title to copyright in SOCI Works A3 and A4, but succeeded regarding SOCI Works A9, A10 and A11. Regarding licensing issues, Sample Licences #2, 3, 10, 11, 13, 30 and 32 were found not to be exclusive licences under section 92 CDPA, whilst Sample Licences #17, #19, #34-38 were determined to be exclusive licences. The court made no finding on the number of Visual Assets or Copyright Works used in training Stable Diffusion, and found no justification for additional damages.

The judgement acknowledged the historic and extremely limited scope of Getty Images' success. The practical impact remains unclear, as the evidence-based approach makes it impossible to determine the scale of real-world watermark generation falling within the established infringement categories. The case highlights the evolving intersection of intellectual property law and artificial intelligence technology, whilst demonstrating the challenges claimants face in establishing comprehensive liability against AI model developers operating in open-source contexts.

Consequential matters arising from the judgement will be addressed at a subsequent hearing.