5 November 2025
Getty Images v Stability AI: the UK High Court decision offers guidance, but critical questions on AI and copyright infringement remain
Case background
The case concerns the AI image generation model, Stable Diffusion, developed and put on the market by the defendant, AI company Stability AI (Defendant).
The claimants include various companies in the Getty Images group (Claimants) who own and manage a massive library of content, including millions of photo, image and video assets, together with identifying captions and key words (Getty Assets). The Getty Assets are hosted on the Claimants’ websites and licensed to its customers on a commercial basis. The Claimants argued that copyright subsists in the Getty Assets, and that its library of Getty Assets constitutes a database.
The Claimants alleged that the training data used by the Defendant to train Stable Diffusion included millions of Getty Assets which had been “scraped”, amounting to the use of the Getty Assets without the Claimants’ consent. Further, the Claimants alleged that some of Stable Diffusion’s outputs (generated in response to user prompts) strongly resembled certain Getty Assets, and in some cases even bore Getty Images trade marks.
The Claimants sought a declaration that the Defendant’s conduct infringed their copyright, trade marks and database rights. They also sought an injunction to restrain the defendant from further infringement, the delivery up of infringing goods and articles, and an inquiry as to damages or an account of profits.
As a brief timeline:
- On 15 January 2023, the Claimants’ claim was issued.
- On 1 December 2023, Smith J of the High Court refused the Defendant’s application for reverse summary judgment on various of the Claimants’ claims (the Summary Judgment Decision, aspects of which are explored below).
- On 14 January 2025, Smith J issued a decision agreeing with the Defendant that the sixth claimant, a picture and video library, could not act as representative for a class of copyright owners, whose works had been exclusively licensed to Getty Images and whose rights the Claimants claimed were also infringed by the Defendant.
- On 9 June 2025, the first trial to determine issues of liability commenced.
- On 4 November 2025, the High Court issued its judgment (the Judgment), given by Smith J.
Copyright infringement at the input stage: the training and development claim
The claim was that the Defendant infringed the Claimants’ copyright in the Getty Assets by downloading and storing the Getty Assets on servers or computers in the UK during the development and training of Stable Diffusion. This amounted to infringement by copying pursuant to section 17 of the Copyright, Designs and Patents Act 1988 (CDPA).
Stability AI applied for reverse summary judgment on this claim, arguing that there was no prima facie evidence that the training and development of Stable Diffusion took place in the UK, and claiming that the computing resources it devoted to the training of Stable Diffusion were all located outside the UK. However, in the Summary Judgment Decision, Smith J refused the application on the basis that there was evidence potentially pointing against the Defendant’s argument, as well as reasonable grounds for believing that disclosure would reveal evidence of where the training and development of Stable Diffusion took place. Here, Smith J took into account that the Defendant was based in the UK, had its principal place of business in the UK and had numerous UK-based employees in its development team, some of whom potentially had accessed cloud computing clusters in their work on Stable Diffusion.
However, prior to and at trial, the Claimants encountered significant difficulty in mounting evidence for their claim, and ultimately they decided to drop it during the course of the trial. As a result, the Judgment does not cover this claim.
Copyright infringement at the output stage: the communication to the public claim
The claim was that the Defendant infringed the Claimants’ right of communication to the public (under section 20 of the CDPA), by making Stable Diffusion available to users in the UK, where Stable Diffusion could be used to generate an image that reproduced the whole or a substantial part of one of the Getty Assets, in which the Claimants enjoyed copyright.
Initially, this claim focused on text prompts, namely prompts identical or similar to the captions used by the Claimants on the Getty Assets. In the Summary Judgment Decision, Smith J allowed the Claimants to expand their case to include image prompts, namely where a user prompts Stable Diffusion with an image that is one of the Getty Assets, applying settings so that the output reproduced the whole or a substantial part of the same. The Defendant had argued that, in the circumstances envisioned in this claim, it was merely acting as intermediary or host. However, Smith J held that whether or not this was factually the case was an issue to be determined at trial.
Similar to the previous claim, the Claimants encountered significant evidentiary challenges in proving their claim, and ultimately dropped it at trial. As a result, the Judgment does not cover this claim.
The secondary copyright infringement claim: importation of an infringing article
The claim was that the Defendant effected secondary infringement of copyright by importing into the UK an infringing article, pursuant to section 22 (importing an infringing copy) and 23 (possessing or dealing with an infringing copy) of the CDPA. The ‘article’ in question here was the pre-trained Stable Diffusion model. If successful, this claim would have prevented the importation and use of the model in the UK, even if all primary acts of infringement (training of the model etc) had taken place outside of the UK.
Stability AI also applied for reverse summary judgment on this claim, arguing that the word ‘article’ for the purposes of the CDPA was limited to tangible things, and did not include things like software made available on a website. In the Summary Judgment Decision, Smith J refused the application, holding that the issue of the correct statutory interpretation of ‘article’ should go to trial, and that this was a novel question with policy ramifications.
In the Judgment, Smith J agreed with the Claimants that ‘article’ was not limited to tangible things and could include software, such as the Stable Diffusion model. However, she held that the pre-trained Stable Diffusion model was not itself an infringing copy. She observed that while the model weights are altered during training by exposure to copyright works in the Getty Assets, the model weights are not themselves an infringing copy, nor do they store or reproduce an infringing copy. By the end of the training process, the model does not store any copyright works in the Getty Assets, nor has it ever done so. Smith J held that it was not relevant that the development of Stable Diffusion separately involved the reproduction of the Getty Assets via local and cloud storage. As a result, Smith J dismissed the Claimants’ claim.
The trade mark infringement and passing off claim
The Claimants’ claim regarding trade marks was that Stable Diffusion can generate outputs displaying the Claimants’ registered trade marks, including “GETTY IMAGES” and “ISTOCK”, in a manner amounting to trade mark infringement pursuant to sections 10(1), 10(2) and 10(3) of the Trade Marks Act 1994, as well as passing off. A representative example was where Stable Diffusion produces an image resembling a photo bearing the standard Getty Images watermark, commonly applied by the Claimants to photo assets in its library.
In the Judgment, the Claimants’ claims were partially successful, but this success was limited. Smith J held that the Claimants had proven trade mark infringement under sections 10(1) and 10(2) for a limited number of images that Stable Diffusion had produced. This involved only historical infringements, for a specific set of images, and did not mean that Stable Diffusions’ outputs more generally occasioned trade mark infringement. Smith J held that infringement under section 10(3) claim was not made out, as there was insufficient evidence that the distinctive character or the repute of the Claimants’ relevant marks had suffered unfair advantage or detriment, nor that the economic behaviour of the average consumer would have been altered by the Defendant’s conduct. Finally, Smith J did not address the claim of passing off, holding that it did not add anything in addition to the trade mark infringement claims already decided.
The database right infringement claims
The Claimants’ claims were based on provisions of the EU Database Directive (Directive 96/9/EC) and the Copyright and Rights in Databases Regulations 1997 (SI 1997/3032), certain effects of which endure in the UK as a result of the UK-EU withdrawal agreement. The Claimants argued that Getty Images enjoys database right in its database of Getty Assets, such database having been arranged in a systematical and methodical way, accessible electronically, and the product of a very significant investment. They claimed that the Defendant infringed this database right by extracting a substantial part of the contents of the database, and re-utilising a substantial part of the contents of the database, in both cases either in one go or in a repeated and systematic manner.
As with the two primary copyright infringement claims above, this claim was ultimately dropped by the Claimants at trial, and is not covered by the Judgment.
Forward outlook
Though the Claimants were successful with some of their trade mark infringement claims, this success was constrained to a very specific set of facts.
Whilst the Judgment on the secondary copyright infringement claim is clear, the most significant claims were by far the primary copyright infringement claims, and particularly the training and development claim. The Claimants’ withdrawal of these claims will disappoint many who were hoping for the court’s guidance on the legality of using third party works to train an AI model.
It may be some time before the primary infringement issues will be resolved in the UK. At least two other cases involving AI and copyright have been threatened in the UK, though it may be years before these are settled or decided. In the meantime, the UK government is considering reforming copyright law to resolve various questions presented by AI, including whether to introduce an EU-style text and data mining exception to infringement (which might cover the “scraping” activity alleged in this case), as well as obligations on AI developers regarding disclosure and transparency regarding training materials used to train AI models. The government’s consultation on this matter closed in February 2025 (see here and here), with more than 11,500 responses received.
As for the Claimants and the Defendant, a parallel suit in the US is ongoing in the Northern District of California, the outcome of which is awaited.
If you have questions regarding anything raised in this article, please get in touch with the authors or your regular DLA Piper contacts.
