Getty Images v. Stability AI: UK High Court draws the line on AI, trademarks, and copyright

The UK High Court’s decision in Getty Images v. Stability AI ([2025] EWHC 2863 (Ch)) highlights the difficulties of reconciling intellectual property law that is decades old with the fast-paced development of generative AI. The case, closely watched by both creative and tech sectors, sets some boundaries for trademark and copyright protection when AI models are trained on third-party works and generate content derived from them. What does this mean for rights holders, AI developers, and anyone using AI-generated content?

The dispute in brief

Getty Images, a global leader in stock photography, brought proceedings against Stability AI, the developer of the “Stable Diffusion” model. Getty alleged that millions of its images – including those bearing its iconic watermarks – were used without consent to train the AI, and that the model could generate synthetic images featuring Getty’s registered trade marks. Claims included trademark infringement, passing off, and secondary copyright infringement.

Key legal issues

• Trademark infringement: Did AI-generated images with Getty watermarks infringe registered trade marks under the Trade Marks Act 1994?
• Secondary copyright infringement: Was the Stable Diffusion model itself an “infringing copy” under the Copyright, Designs and Patents Act 1988, given its training on copyright works (albeit this training took place outside the UK)?
• Passing off: Did the generation of images with Getty’s watermarks amount to actionable misrepresentation?

The court’s findings

Trademark infringement

The court found that certain versions of Stable Diffusion, when accessed via certain platforms (DreamStudio and the Developer Platform), could generate images with watermarks identical or highly similar to Getty’s and iStock’s registered marks. This amounted to trademark infringement under sections 10(1) and 10(2) of the TMA, but only for specific instances and not for later versions (such as SD XL and v1.6). The court emphasised that infringement depended on the clarity and recognisability of the synthetic watermark and did not extend to “garbled” or distorted signs. It rejected claims under section 10(3) (reputation) due to insufficient evidence of detriment to the distinctive character or reputation of the marks, or of unfair advantage.

Secondary copyright infringement

Getty’s claim that the AI model itself was an “infringing copy” was dismissed. The judge held that, under UK law, an “infringing copy” must contain an infringing copy of the original work. Since Stable Diffusion’s model weights (the learned instructions for generating new images) did not store or reproduce the original images, and the training occurred outside the UK, there was no infringement.

Passing off

The court declined to address passing off in detail, noting that the outcome would not differ from the trademark analysis.

Remedies

Getty succeeded only in part, with findings limited to specific instances of trademark infringement. Claims for additional damages and secondary copyright infringement were dismissed.

What’s new or unusual?

• This is the first UK decision to address trademark infringement by AI-generated synthetic images, focusing on watermarks as a source of confusion.
• The judgment clarifies that “infringing copy” under UK copyright law does not extend to AI models that do not store or reproduce the original works, even if trained on them.
• The case highlights the evidential challenges in proving real-world confusion or infringement by AI outputs, especially where outputs are ephemeral or technical steps (like watermark filtering) can mitigate risk.

Practical takeaways

• For rights holders: Trademark owners can succeed against AI developers where synthetic outputs reproduce registered marks in a way likely to cause confusion—but only if the evidence is clear and not “contrived.” Copyright claims against AI models trained outside the UK face significant hurdles.
• For AI developers: Technical measures, such as watermark filtering, can be effective in limiting liability. The court was cautious not to extend liability for AI-generated outputs beyond what the evidence and statutory language support.
• For the industry: The case leaves unresolved the “burning issue” of whether training Stability Diffusion would have amounted to primary infringement of copyright in the UK had training taken place here.

Conclusion

Getty Images v. Stability AI is essential reading for anyone advising on IP and AI. It demonstrates the importance of clear evidence, technical understanding, and strategic litigation choices in this fast-evolving area. For rights holders, it’s a reminder to review licensing, watermarking, and enforcement strategies. For AI developers, it’s a call to invest in robust filtering and compliance measures.

← Back to News