Stability AI defeats Getty Images’ copyright claims in first of its kind dispute before the High Court in London

On 4 November 2025, UK High Court Judge Joanna Smith DBE handed down her much-anticipated judgment in Getty Images v Stability AI. Following Getty’s abandonment of their primary copyright claims during trial, Stability has now prevailed on the remaining secondary copyright infringement issue. The judgment also addresses Getty’s trade mark infringement claims, dismissing the claim under section 10(3) of the Trade Marks Act while reaching “extremely limited” infringement findings under sections 10(1) and 10(2) for early versions of Stable Diffusion.     

As one of the first IP claims against an AI developer to reach trial, the case has inevitably attracted significant public interest. In January 2023, Getty filed a broad claim encompassing copyright, database right, trade mark infringement and passing off allegations relating to the Stable Diffusion model itself, its training and development process and images generated by it. 

By the end of trial in June 2025, only two claims remained. The first was whether Stability’s making the model weights for certain versions of Stable Diffusion available for download via Hugging Face was a secondary infringement of Getty’s copyright. The second was whether Stability had infringed either the Getty or iStock trade marks due to the presence of watermark-like features in certain Stable Diffusion outputs identified by Getty.

The parties

Founded in 1995, Getty (which also owns the iStock brand) is a global visual content creator and marketplace which licenses rights to editorial and stock photographs and videos. This includes copyright works and other visual assets owned by Getty, along with third party content. Getty store their content in a database together with associated metadata and offer it to the public to license via Getty and iStock branded websites.

Stability is a leading AI company, describing itself as the enterprise-ready creative partner for teams and creators. It delivers professional-grade generative AI tools and solutions for media generation and editing across image, video, 3D, and audio. These include image generation models, released under the name Stable Diffusion. One of the first text-to-image models, Stable Diffusion version 1, was released by researchers in the CompVis group at Ludwig Maximilian University of Munich in August 2022. Work on later versions of Stable Diffusion shifted to Stability, which released Stable Diffusion v2, Stable Diffusion XL/1.6 and more recent versions, including Stable Diffusion 3.0 and 3.5. 

The claims

Getty’s claim alleged that URLs to Getty’s images were included in the LAION 5B dataset, referred to as a source of training data in the model cards for Stable Diffusion v1 and v2. Getty argued that by using their content when creating Stable Diffusion, Stability had infringed Getty’s copyright and database right. They alleged that 7.3 million Getty works were used to create Stable Diffusion v1 and 4.4 million for v2. Getty also argued that a consequence of using their copyright works in training was that Stable Diffusion itself was an article which was an infringing copy of Getty’s copyright works.

Getty further alleged that in some cases, outputs from Stable Diffusion infringed Getty’s copyright, database right and/or trade marks. While the initial focus was on Stable Diffusion v1 and v2, Getty expanded their claim in the summer of 2024 to include two later versions: Stable Diffusion XL and 1.6 (a version based on the model architecture of XL, rather than earlier v1 models). Getty did not make any infringement allegations regarding more recent versions of Stable Diffusion, including versions 3.0 and 3.5.

In response, Stability argued that the development and training of Stable Diffusion took place outside the UK. As Getty’s claim was limited to their UK copyright and database right, Stability argued that those rights had not been engaged by the process of making Stable Diffusion. Stability also challenged Getty’s assertion that outputs from Stable Diffusion could infringe Getty’s copyright, database right and/or trade marks, or that Stability had committed any act of secondary copyright infringement by dealing with the models in the UK.

Intensive case management as Getty’s case narrows

Over the course of ten interim hearings (all before the trial judge), three areas of case management impacted the scope of Getty’s claim: (1) the scope of Getty’s copyright and database right outputs claim; (2) the representative claim pursued on behalf of over 50,000 photographers and contributors who were said to exclusively license their content to Getty; and (3) the scope of Getty’s pleading regarding the training and development claim.

(1) Getty’s outputs case

Following a CMC in July 2024, Getty served a Statement of Case of Infringement identifying 17 copyright works and just over 200 outputs generated by Getty from Stable Diffusion (SOCI).  While these outputs were first said to be ‘representative examples’ of copyright and database right infringement, when pressed by Stability to explain how they had been selected, Getty narrowed their outputs claim to only the 17 specific copyright works identified in their SOCI (out of the millions they alleged were used to train Stable Diffusion). 

The claim narrowed further to cover (i) only the specific text prompts Getty had used to obtain the SOCI outputs (which were either taken verbatim from captions used on the Getty website to describe specific Getty works, or reworded versions of these which their expert had created using ChatGPT) and (ii) only the 200 or so specific outputs in Getty’s SOCI. By trial, the number of copyright works Getty relied on had reduced from 17 to 13, due to title issues that Getty faced.

Following the close of evidence at trial, Getty abandoned all of their copyright and database rights claims relating to Stable Diffusion’s outputs.

(2) The representative claim

Getty’s business model includes offering content owned by third parties under licence agreements.  Getty sought to have one of these third party licensors, Thomas M. Barwick, Inc, act as a representative claimant, on behalf of a class of 50,000 other exclusive licensors of rights that Getty alleged were also infringed by Stability.

Stability opposed the representative claim, arguing that the claims of the members of the purported class did not raise a common issue. In a judgment handed down on 14 January 2025 ([2025] EWHC 38 (Ch)), the judge rejected Getty’s representative claim, finding that to decide whether each licensor fell within the proposed class would require an individualised assessment of whether their copyright works has been used to train Stable Diffusion. 

Following this decision, Stability agreed that Getty could proceed with their own claims relating to the copyright works that they exclusively licensed without joining every licensor to the proceedings subject to Getty indemnifying Stability against subsequent claims. This set up a complex dispute about which of Getty’s standard form contributor agreements met the requirements for an exclusive licence under section 92 of the Copyright, Designs and Patents Act (CDPA).

(3) Getty’s case on training and development

Getty’s claim that Stability had infringed their UK copyright and database right while training and developing Stable Diffusion was pleaded as an inferential case. Getty said that Stability downloaded images in the UK by reason of certain factors such as being a UK registered company and employing developers in the UK. This inferential case survived Stability’s application for summary judgment in 2023 ([2023] EWHC 3090 (Ch)) but remained unparticularised until seven weeks before trial when Getty sought permission to introduce a 28-page Statement of Case on Development and Training (SOCDT). The SOCDT was said by Getty to clarify their existing pleading on location, rather than constitute a late amendment. In a judgment given on 1 May 2025, the judge agreed with Stability that the SOCDT was a very late amendment and that Stability would suffer significant prejudice if the full amendment was allowed ([2025] EWHC 109 (Ch)).

Following the close of evidence at trial, Getty abandoned all of their copyright and database right claims regarding the development and training of Stable Diffusion.

The judgment

Following the narrowing of Getty’s claim, the key issues which remained for determination in the judgment were:

• Secondary copyright infringement: By making the model weights for versions 1, 2, XL and 1.6 of Stable Diffusion available for download via Hugging Face, had Stability committed secondary copyright infringement under sections 22 or 23 CDPA?

• Trade mark infringement: Had Stability infringed the Getty or iStock trade marks under sections 10(1), 10(2) or 10(3) Trade Marks Act (TMA) due to the presence of watermark-like features in certain Stable Diffusion outputs identified by Getty?

Secondary copyright infringement

Getty’s secondary infringement claim required a finding that Stable Diffusion v1, v2, XL or 1.6 was an article which is an “infringing copy” because it had been imported into the UK and its making in the UK would have constituted an infringement of the copyright in a Getty copyright work, pursuant to section 27(3) CDPA. This gave rise to two novel questions of law: whether the model weights files for Stable Diffusion were capable of being (i) an “article” for the purposes of sections 22 and 23 CDPA and (ii) an “infringing copy” for the purposes of section 27 CDPA.

On the first issue, the judge concluded that the definition of an article could extend beyond tangible articles and include electronic copies stored in an intangible form. On the second issue, Getty had accepted (in line with evidence from both parties’ experts) that the Stable Diffusion model weights do not contain copies of any Getty copyright works. However, Getty argued that all section 27(3) requires is that the making of the article would have constituted an infringement if done in the UK. The judge disagreed, holding that an article which is an “infringing copy” must have at some point in time consisted of, contained, or stored a copy of a copyright work, which was not the case here. The judge therefore dismissed Getty’s secondary infringement claim.

Trade mark infringement

Getty pursued a claim for infringement of Getty and iStock trade marks under sections 10(1), 10(2) and 10(3) TMA for Stable Diffusion versions 1, 2, XL and 1.6. In support, Getty relied on examples of Stable Diffusion outputs containing what they said were versions of the Getty and iStock watermarks. These fell into two categories: (i) examples generated by Getty’s representatives through litigation experiments conducted for the purposes of the proceedings and (ii) some limited examples that Getty and their legal representatives had found “in the wild”, generated by third parties.

Getty’s own examples had been generated using four categories of text prompt. The first two were text prompts taken verbatim from image captions on the Getty website, and reworded versions of those captions created using ChatGPT. The third was text prompts created by one of Getty’s solicitors, loosely inspired by other Getty captions and following the ‘house style’ for Getty’s captions on news, sport, entertainment and contemporary content. The fourth category was text prompts invented by one of Getty’s solicitors which contained either the phrase “news photo” or “vector art”.

Stability argued at trial that Getty had been unable to prove that any UK user of the various Stable Diffusion models in issue had ever been presented with a watermark-like feature on a synthetic image in the UK. Stability pointed to the lack of any real-world evidence that Stable Diffusion users would use Getty captions (verbatim or reworded) as text prompts. This was supported by the results of a prompts sample produced by Stability which suggested that Stable Diffusion users did not use Getty captions as prompts. The judge agreed, finding no evidence that a single user in the UK had copied and pasted a verbatim Getty caption as a text prompt for Stable Diffusion, much less that they had generated a watermark as a result. Reviewing the evidence, the judge also reached the same conclusion for reworded Getty captions and prompts in the style of Getty captions.

As the only evidence advanced by Getty for watermark generation by Stable Diffusion XL and 1.6 came from their own experiments (with no evidence in “the wild”) and used verbatim or reworded Getty captions, the judge dismissed all trade mark claims against these versions. 

For versions 1 and 2 of Stable Diffusion, the judge concluded that, on balance, at least one user in the UK would have generated a synthetic image containing a Getty or iStock watermark. For the purpose of her analysis under sections 10(1) and 10(2), the judge then went on to compare the Getty and iStock marks with examples of watermarks that Getty asked her to consider. In doing so, the judge identified two examples of images generated using Stable Diffusion v1 which she considered to give rise to trade mark infringement of the iStock mark under section 10(1) or 10(2). She also identified one example of an image generated using Stable Diffusion v2 which she considered to give rise to trade mark infringement of the Getty mark under section 10(2). Given the very small number of examples presented to the judge, and the lack of any evidence which would allow her to reach any conclusion regarding the frequency with which watermarks appeared which were sufficiently similar to the marks as to infringe, she summarised her infringement findings as “extremely limited”. 

Given this very narrow finding, the Judge went on to dismiss the infringement claim under section 10(3), citing a lack of evidence that the Getty or iStock marks would suffer one of the relevant categories of harm or that there would be a change in the economic behaviour of the average consumer.   

Other issues

Aside from secondary infringement and trade mark infringement, the judgment also addresses:

Passing off: The judge did not see that this claim added anything to her trade mark findings and declined to address it, on the basis that that certain points had not been properly argued before her.

• Responsibility for release of Stable Diffusion v1 model weights: The judge agreed with Stability that it did not bear any direct liability for any tortious acts alleged by Getty arising from the release of Stable Diffusion v1 via the CompVis GitHub and CompVis Hugging Face pages.

• Title issues: The judge held that seven out of the 14 sample licences relied on by Getty fell within the definition of an exclusive licence under section 92 CDPA and seven did not; and that Getty had made out title to three of the five disputed works relied on in the SOCI but had failed on the other two, leaving eleven of the original seventeen standing.

• Number of works used in training: The judge declined Getty’s request to reach a finding on how many Getty copyright works were used to train Stable Diffusion, considering it unnecessary and not a conclusion she was able to reach on the evidence before her.

• Additional damages: Getty argued that additional damages should be awarded against Stability, saying it had engaged in infringement of “numerous different intellectual property rights on a blatant and widespread scale”. The judge rejected this, referring to her finding that she was unable to find widespread instances of trade mark infringement, or that they continued beyond the release of Stable Diffusion v2. In the circumstances, the judge concluded that there was no basis for a claim for additional damages.

You can read the full High Court judgment here. 

Bird & Bird acted for Stability AI. The team was led by partners Toby Bond and Nick Aries, together with team members Louise O’Hara, Aneesah Kabba-Kamara, William Wortley, Zoe Fuller, Sam Triggs and Tristan Sherliker.