United Kingdom - Getty Images (US) Inc & Ors v Stability AI Ltd, 16-17 December 2025

Claimants: Getty Images (US) Inc & Ors (‘Getty’)

Defendant: Stability AI Ltd (‘Stability AI’)

The Court handed down its judgment following the trial heard between 9–30 June 2025 before Mrs Justice Joanna Smith in the High Court, after Stability’s earlier attempts to secure summary judgment in respect of various of Getty’s claims was dismissed by the Courts.

Getty advanced four claims: (i) copyright infringement, (ii) database right infringement, (iii) trade mark infringement, and (iv) passing off, all as summarised below.

At trial, Stability defended Getty’s claims by arguing, amongst other things, that (a) any training of its gen-AI tool had taken place outside of the UK (negating jurisdiction), (b) it had not reproduced substantial parts of Getty’s works, (c) Getty did not command any database rights due to an insufficient investment in its image library and non-qualifying ownership, and (d) Getty’s trade mark and passing off claims were bound to fail because there had been no active use of Getty’s marks or use of any such marks by Stability “in the course of trade” (and that any of Getty’s watermarks appearing in Stable Diffusion ‘output’  was due to “wilful contrivance” on the part of Getty).

Late in the trial Getty dropped its primary copyright and database right infringement claims, citing evidential weaknesses and jurisdictional hurdles which it had failed to overcome following witness examination and pre-trial disclosure, however, in her Order sealed on 5 January 2026, Mrs Justice Joanna Smith ruled that these claims were “dismissed in circumstances where those claims were fought to closing submissions and then abandoned”. This essentially left only three limbs to Getty’s claims – secondary copyright infringement, trade mark infringement, and passing off. The most interesting of these arguments is the first – with Getty arguing that Stability’s AI tool itself was an “article” for the purposes of s22 CDPA 1988.

On 4 November 2025, the High Court issued judgment. The Court rejected Getty’s secondary copyright claim in full, finding that although an AI model may in principle constitute an “article” under s.22 CDPA 1988, Stable Diffusion did not embody any infringing copies of Getty works. The Court also rejected Getty’s reputation-based trade mark claims and made no determination on passing off. Getty succeeded only on a narrow set of trade mark issues relating to isolated, historic reproductions of ISTOCK and (for v2.1) GETTY IMAGES watermarks. Stability succeeded on the vast majority of the issues and was granted leave to appeal the secondary infringement claim on 17 December 2025.

This judgment represents the first detailed judicial assessment in the UK of secondary copyright infringement in the context of AI model weights, and it significantly limits the circumstances in which AI developers may incur liability for model architecture or distribution.

Getty brought proceedings against Stability AI in January 2023. The legal arguments surrounded the use/reproduction of images from Getty’s websites to train Stability AI’s generative AI tool, “Stable Diffusion”. Stable Diffusion is a deep-learning, text-to-image (and now image-to-image) AI model that is used to generate synthetic images in response to commands entered by users.  

Getty claimed that the first iteration of Stable Diffusion was trained using 12m photos, videos and illustrations from Getty’s websites (of which 7.3m are copyright works). It also claims the second iteration was trained using 7.5m of Getty’s photos, videos and illustrations (of which 4.4m are copyright works).  

Stability AI made its tool available to users in the UK on its website, Dream Studio, where it sells access to compute time for generating images with Stable Diffusion. Stable Diffusion is also available to third parties in the UK to download on an open-source basis.

Getty was represented by Fieldfisher and Stability AI was represented by Bird & Bird.

Copyright

The Court held that Getty’s remaining copyright claim - secondary infringement under ss.22–23 CDPA - failed in its entirety. In particular, the Court determined that:

• Although the word “article” in ss.22–23 CDPA is capable of extending to intangible items, including digital model weights distributed electronically, Stable Diffusion does not constitute an “infringing copy” of any Getty work. The Judge found that the model did not store or embody reproductions of whole or substantial parts of any Getty copyright works, whether in their original or transformed (e.g., “noisy”) form.
• Getty did not demonstrate that the weights contained any stored images, compressed copies, or other fixed reproductions of any specific Getty work. As a result, the Court held that the model could not meet the statutory requirement of being “an article” that is an infringing copy of a protected work.
• The Court emphasised that Stable Diffusion’s architecture and weights operate at a high level of mathematical abstraction that does not encode, contain, or allow extraction of any specific image, and therefore do not amount to stored copies or substantial parts of any Getty work.
• Because Getty’s primary and output-based copyright claims had already been abandoned at trial for want of evidence and jurisdiction, and because the model itself was not an infringing article, Getty’s secondary infringement claim failed and no liability arose under ss.22–23 CDPA.

Trade Marks

Getty succeeded only on a narrow and historically confined aspect of its trade mark claim. The Court found as follows:

• Getty established a limited infringement under s.10(1) TMA (double identity) in relation to the ISTOCK mark, but only in respect of early v1.x iterations of Stable Diffusion and only because Getty demonstrated one real-world instance in which an AI-generated image contained an identical ISTOCK watermark (“the Spaceships Image”). The Court noted that this example involved a precise reproduction of the ISTOCK sign and therefore satisfied the identity requirement.
• For the GETTY IMAGES mark, the Court found no s.10(1) infringement, as none of the examples relied upon by Getty constituted an identical reproduction of the mark.
  • Under s.10(2) TMA (likelihood of confusion), Getty succeeded in relation to:
  • the ISTOCK mark for v1.x outputs, and
  • the GETTY IMAGES mark for v2.1 outputs, where the Court held that certain synthetic images contained watermarks that were sufficiently close to the respective marks to give rise to a likelihood of confusion.
• The Court rejected all claims under s.10(3) TMA (reputation), finding no persuasive evidence of detriment, unfair advantage, dilution or tarnishment arising from the isolated outputs identified.
• The Judge further held that there was no evidence of infringing use in the UK for later models (including SD v1.6 and SD XL), because Getty could not show that the synthetic images relying on blurred or distorted watermarks had been produced in the UK in the ordinary course of trade.

Passing Off

• The Court declined to rule on the passing-off claim. Getty’s passing-off allegations had been tied to the same instances of watermark reproduction relied upon for the trade mark claims. Given the very limited findings of trade mark infringement and the Court’s view that passing off added no materially distinct issues, the Judge considered it unnecessary to decide the claim.

For expert analysis on the case see:

• Getty Images v Stability AI - Getty unsuccessful in secondary copyright infringement claim (re stable diffusion)
• Getty Images v Stability AI - Getty's narrow win in trade mark infringement claim