Getty Images v Stability AI: A landmark IP battle in the age of Generative AI

Getty Images v Stability AI is a landmark case testing how copyright and trade mark law apply to generative AI.

The judgment, handed down on 4 November 2025, delivers a mixed outcome that highlights the limits of existing IP frameworks and sets important precedents for both rights holders and AI developers.

This has been one of the most closely watched intellectual property cases in the age of generative AI. The dispute pits Getty Images – custodian of a vast archive of photographic works – against Stability AI, developer of the Stable Diffusion image generator. At stake were foundational questions about how copyright and trade mark law apply to modern AI systems trained on large datasets. Getty Images initially advanced a broad suite of claims, including primary and secondary copyright infringement, database right infringement, trade mark infringement, and passing off. However, shortly before closing submissions, Getty Images withdrew several key claims, narrowing the scope of the litigation and leaving unresolved issues of importance to both the creative and technology sectors.

Background

Getty issued proceedings against Stability AI in January 2023, prior to the launch of Stable Diffusion XL. The litigation was marked by extensive case management and complex procedural ’firsts’, reflecting the novelty of the issues raised. On 22 April 2024, the Court ordered a first trial on liability, recognising the need to resolve foundational legal questions before addressing remedies. In the six months leading up to trial, the Court held frequent hearings, many resulting in reported judgments that shaped the scope of the dispute.

By the time of trial, Getty had narrowed its case significantly, withdrawing several claims and focusing on trade mark infringement, secondary copyright infringement, and licensing issues. Broadly:

(a) Whether the normal use of Stable Diffusion in the UK generates synthetic images bearing Getty Images’ trade marks, amounting to trade mark infringement or passing off.

(b) Whether Stability AI committed acts of secondary copyright infringement by importing, possessing, or distributing Stable Diffusion as an “infringing copy”.

The Judgment

The court handed down judgment on 4 November 2025. It was a mixed outcome (largely in Stability Diffusion’s favour) that reflects the legal and technical complexity of the case.

Trade Mark Infringement

Getty succeeded in part under sections 10(1) (double identity) and 10(2) (likelihood of confusion) of the Trade Marks Act 1994. The court found that early versions of Stable Diffusion (v1.x and v2.x), when accessed via DreamStudio or the Developer Platform, had generated synthetic images containing Getty and iStock watermarks, amounting to trade mark infringement. However, the findings were described as “historic and extremely limited in scope”, with no evidence of infringement in later versions such as SD XL or v1.6.

The s.10(3) claim on grounds of dilution, tarnishment and unfair advantage was dismissed for lack of evidence of harm or change in consumer behaviour. The court ultimately declined to rule on passing off.

The court emphasised that infringement depended on real-world use, not “contrived” prompts, and that Getty had succeeded only where it could show actual user-generated outputs containing watermarks. Importantly, the Court placed responsibility on the model provider rather than end-users for trade mark bearing outputs, a point Getty underscored in its public statement.

Copyright Infringement

Getty’s primary copyright claims were withdrawn during trial due to jurisdictional limitations – the training of Stable Diffusion occurred outside the UK. The remaining secondary infringement claim was dismissed. The court held that Stable Diffusion, while trained on Getty’s content, did not store or reproduce any copyright works and therefore there was no “infringing copy”.

This finding reinforces a critical distinction in AI copyright litigation: the difference between pattern learning and data reproduction. It sets a high bar for future claims based on training data alone, particularly where models do not retain or regenerate identifiable works.

Other Issues

Getty succeeded in establishing ownership of some sample works and exclusive rights under certain licences, but failed on others. The court declined to make findings on the number of visual assets used in training and found no basis for additional damages.

Why This Case Matters

The implications of this decision are far-reaching. It exposes the limitations of existing copyright frameworks in addressing AI models trained on vast datasets, especially where training takes place outside of the UK.

While the judgment acknowledges potential trade mark risks in generative AI outputs, it sets a high evidentiary bar for future claims. More broadly, the case underscores the tension between protecting creative industries and enabling technological innovation. Courts, the judgment makes clear, will not resolve these policy questions without legislative intervention. For now, rights holders may need to rely more heavily on contractual and licensing strategies, while regulators consider whether bespoke rules for AI training and data use are required to strike a fair balance between creators and innovators.

Getty is pursuing related claims in the United States, where different doctrines (e.g., fair use) and procedures may test questions that this judgment left open.

Practical Takeaways for Rights Holders & AI Developers

• Rights holders: Consider contractual/licensing strategies, watermark detection pipelines, and consider assertive trade mark enforcement where outputs display branding; gather user-generated evidence from the relevant jurisdiction.
• Developers: Maintain output safeguards (e.g., watermark blocking/filters), keep versioned change logs, and document training locations and data governance to manage jurisdictional exposure.

If you would like to discuss any of the issues raised in this article please contact Emily Roberts. Chloe Perea Poole or another member of the IP team.

Subscribe to our Concept newsletter and receive the latest intellectual property legal updates, news and event invitations direct to your inbox.