Last reviewed
May 19, 2026
Sources
No citations yet
Review status
Needs citations
Revision
v1 · 4,252 words
Add missing citations, update stale details, or suggest a clearer explanation.
Getty Images v. Stability AI refers to a pair of parallel intellectual property lawsuits brought by visual content licensing company Getty Images against Stability AI, the British developer of the Stable Diffusion text-to-image generative AI model. The cases concern allegations that Stability AI used millions of Getty's copyrighted images, captions, and metadata, scraped from Getty's websites without authorisation, to train Stable Diffusion. Getty issued UK High Court proceedings in London on 16 January 2023 and filed a parallel complaint in the U.S. District Court for the District of Delaware on 3 February 2023.[^1][^2][^3]
After a three-week trial in June 2025 and a 219-page judgment handed down on 4 November 2025, Mrs Justice Joanna Smith DBE largely rejected Getty's claims in the UK, finding only a "historic and extremely limited" trade mark infringement and dismissing Getty's secondary copyright infringement claim outright. Getty had abandoned its primary copyright and database right claims shortly before closing submissions after accepting that there was no evidence that Stable Diffusion was trained in the UK.[^4][^5][^6] In the United States, Getty voluntarily dismissed its Delaware action on 14 August 2025 and refiled the same day in the U.S. District Court for the Northern District of California, where the case remains pending as of May 2026.[^7][^8]
The dispute is widely regarded as the first substantive judicial test in the English courts of how UK copyright and trade mark law apply to the development of generative AI systems, and one of the highest-profile rights-holder versus AI developer disputes globally.[^4][^9] No settlement had been reached as of May 2026, and both parties have signalled an intention to pursue the appellate process.[^10]
Stable Diffusion is an open-source text-to-image diffusion model first released publicly in August 2022 by Stability AI in collaboration with researchers at LMU Munich and Heidelberg University. It generates images in response to text or image prompts and was made available for download as well as through Stability AI's commercial interfaces, including a paid platform marketed as DreamStudio.[^9][^11]
The model was trained on large-scale image-text datasets compiled by LAION, a German non-profit research association. LAION's datasets, notably LAION-5B, do not host images themselves but consist of URLs to images scraped from the public web together with associated captions and metadata. Getty Images alleged in its UK and US pleadings that more than 12 million of its copyright images, along with associated captions and metadata, were referenced in LAION-5B and used by Stability AI to train successive versions of Stable Diffusion without Getty's consent.[^1][^2][^11]
Getty Images is one of the world's largest visual content suppliers, operating both the Getty Images and iStock licensing platforms. Both brands attach distinctive watermarks to preview images displayed on their websites, and both have registered UK and EU trade marks covering those watermarks. Getty itself licenses imagery to artificial intelligence developers for model training under what it has publicly described as "viable licensing options" that "respect personal and intellectual property rights," and in 2023 launched its own image-generation product trained on its proprietary library.[^1][^9]
Getty Images issued a public statement on 17 January 2023 announcing that it had commenced UK High Court proceedings against Stability AI. The statement asserted that "Stability AI unlawfully copied and processed millions of images protected by copyright and the associated metadata owned or represented by Getty Images absent a license" and added that "Stability AI did not seek any such license from Getty Images and instead, we believe, chose to ignore viable licensing options and long-standing legal protections."[^1] Stability AI's then chief executive Emad Mostaque, who co-founded the company in 2020 and resigned as CEO in March 2024, publicly maintained that the way generative AI transforms training material qualifies it as "fair use" under US law.[^12][^13]
Getty Images issued proceedings on 16 January 2023 in the High Court of Justice of England and Wales, Business and Property Courts, Intellectual Property List (Chancery Division). The six claimants were Getty Images (US) Inc, Getty Images International U.C., Getty Images (UK) Limited, Getty Images Devco UK Limited, iStockphoto LP and Thomas M. Barwick, Inc, and the sole defendant was Stability AI Limited, the English-incorporated entity in the Stability AI group.[^3][^14]
Getty's pleaded case in the UK was broad. It included five distinct heads of claim: (i) primary copyright infringement based on web-scraping and on the alleged storage of training data and model weights in the UK; (ii) secondary copyright infringement on the basis that Stable Diffusion was an "infringing copy" the importation or dealing in of which infringed copyright; (iii) database right infringement; (iv) trade mark infringement under sections 10(1), 10(2) and 10(3) of the Trade Marks Act 1994, principally on the basis that Stable Diffusion outputs in some cases reproduced the Getty Images and iStock watermarks; and (v) passing off.[^4][^5][^9]
In December 2023 the High Court refused Stability AI's application for summary judgment and reverse summary judgment, ruling that the case could proceed to trial. The judge at the interlocutory stage held that Getty's claims raised triable issues of fact and law and were not suitable for summary disposal.[^15]
On 25 June 2025, during the trial itself, Getty Images dropped its primary copyright infringement claim relating to model training, accepting that there was no evidence that training had taken place within the United Kingdom. Getty also abandoned its database right claim. The trial therefore proceeded only on the secondary copyright infringement claim, the trade mark claims and the passing off claim.[^4][^5][^16]
The liability trial took place between 9 June 2025 and 30 June 2025 in the High Court before Mrs Justice Joanna Smith DBE. The court heard evidence from 19 witnesses, including four witness statements from Getty Images executive Ms Cameron, who gave evidence on the Getty Images database, watermarks, licensing practices and an analysis of the Getty works identified in the LAION datasets. Combined skeleton arguments from the parties ran to over 560 pages.[^4][^17]
Stability AI's defence ran along several main lines. First, it argued that all training of Stable Diffusion took place outside the UK, principally on Amazon Web Services compute resources in the United States, and accordingly fell outside the territorial reach of the UK Copyright, Designs and Patents Act 1988 (CDPA). Second, it argued that Stable Diffusion model weights are not stored copies of any training image but rather statistical parameters learned during training, so the model cannot be an "infringing copy" within the meaning of section 27 CDPA. Third, on trade mark, Stability AI argued that any watermark reproductions in outputs were the result of users entering "atypical" prompts and that there was no use of the marks "in the course of trade" by Stability AI itself.[^4][^11][^17]
On 3 February 2023, Getty Images (US), Inc. filed a complaint against Stability AI, Inc. in the United States District Court for the District of Delaware. The case was assigned case number 1:23-cv-00135. Getty filed an Amended Complaint on 29 March 2023 adding Stability AI Limited (the UK entity) as a co-defendant, and a Second Amended Complaint on 9 July 2024 adding a third defendant, Stability AI US Services Corporation.[^2][^7]
The Delaware complaint alleged copyright infringement, providing false copyright management information, removal or alteration of copyright management information under the Digital Millennium Copyright Act (DMCA), federal and state trade mark infringement, trade mark dilution, unfair competition and deceptive trade practices. Getty alleged that Stability AI had reproduced "more than 12 million" of its photographs along with captions and metadata in connection with training and operating Stable Diffusion and the DreamStudio interface.[^2][^18][^19]
Public reporting indicated that Getty was seeking up to approximately $1.7 billion in damages in the US, reflecting statutory damages of $150,000 for each of 11,383 works for which Getty was pursuing claims in the litigation. Getty's chief executive Craig Peters told reporters in 2024 that the company was "spending millions and millions of dollars in one court case," describing the litigation as "extraordinarily expensive."[^18][^20]
Stability AI moved to dismiss the original complaint in May 2023, challenging the Delaware court's personal jurisdiction over both the US and UK Stability entities and seeking, in the alternative, transfer of the case to the Northern District of California where parallel artist class-action proceedings were already pending. After Getty filed its Second Amended Complaint in July 2024, Stability AI renewed its motion to dismiss, again challenging personal jurisdiction over Stability AI Limited and again seeking transfer to California.[^7][^21]
Rather than wait for the Delaware court to rule on Stability AI's renewed motion, Getty Images voluntarily dismissed its Delaware case on 14 August 2025. The Delaware docket was formally terminated on 18 August 2025.[^7][^8]
On the same day as the dismissal, 14 August 2025, Getty refiled substantially the same claims in the U.S. District Court for the Northern District of California. The new action, Getty Images (US), Inc. v. Stability AI, Ltd. et al., case number 3:25-cv-06891, was assigned to United States Magistrate Judge Lisa J. Cisneros. Reporting at the time noted that Getty's new complaint added a copyright dilution theory, alleging that Stability AI's alleged unlicensed use had caused "dilution of the market for Getty Images' copyrighted works," and reiterated the DMCA, trade mark and unfair competition theories from the Delaware action. Getty has continued to seek a permanent injunction against further use and distribution of Stable Diffusion models trained on Getty content, together with damages.[^8][^22]
As of May 2026, the California action remains in early-stage proceedings. No trial date had been publicly set, and there had been no settlement between the parties.[^8]
Mrs Justice Joanna Smith DBE handed down judgment on 4 November 2025. The judgment, Getty Images (US) Inc & Ors v Stability AI Limited [2025] EWHC 2863 (Ch), ran to 219 pages and addressed each of the surviving claims in turn.[^14][^17]
The central remaining copyright claim was that Stable Diffusion itself was a "secondary infringement" article: an article whose making had involved the use of infringing copies of Getty works during training, and which Stability AI then imported into or distributed in the UK. Sections 22 and 23 of the CDPA make it an act of secondary infringement to import, possess, sell or distribute an article that is or that the importer or distributor knows or has reason to believe is an "infringing copy" of a copyright work.[^4][^5][^17]
Justice Smith made one finding favourable to Getty as a matter of legal principle: she held, for the first time in English case law, that an "article" for the purposes of sections 22 and 23 CDPA is not limited to a tangible physical object, and can include intangible items such as an AI model.[^4][^5] However, she went on to reject the substantive claim on the basis that Stable Diffusion is not an "infringing copy." She wrote that the model weights "are purely the product of the patterns and features they have learnt over time during the training process" and contain no copies of Getty's works. As she put it, the question was "whether an article whose making involves the use of infringing copies, but which never contains or stores those copies, is itself an infringing copy such that its making in the UK would have constituted an infringement... In my judgment, it is not."[^6][^17]
Because the model does not "contain, or [has not] at some point contained" a copy of any Getty work, dealings with the model in the UK could not amount to secondary infringement, regardless of what occurred during training abroad. The secondary copyright claim was accordingly dismissed.[^4][^5][^6]
Getty's trade mark claims focused on the appearance of Getty Images and iStock watermarks in some Stable Diffusion outputs. Getty sued under three provisions of the Trade Marks Act 1994: section 10(1) (identical mark, identical goods/services), section 10(2) (similar mark causing likelihood of confusion) and section 10(3) (mark with reputation, harm without due cause).
Justice Smith found that certain early versions of Stable Diffusion (the v.1.x series), when accessed through DreamStudio and Stability AI's developer platform, were capable of producing outputs that included signs identical or confusingly similar to Getty's and iStock's registered watermark marks. She found "extremely limited" infringement of iStock marks under sections 10(1) and 10(2) for those early models. She rejected Getty's section 10(3) dilution claim in full, holding that there was insufficient evidence of real-world economic harm or change in average consumer behaviour, and characterising parts of Getty's dilution case as "pure supposition." For later versions of Stable Diffusion (v.2.x, SD 1.6, SD XL), she found "not one jot of evidence" of real-world watermark generation by UK users and accordingly no infringement.[^4][^5][^17][^23]
Importantly for the broader policy question, the court did not accept Stability AI's argument that any watermark output was solely the responsibility of users. The judge accepted the principle that a model provider that "choose[s] to train the models on images bearing Getty's trade marks" cannot avoid liability by attributing all use to end users.[^17][^23]
Justice Smith declined to make a substantive ruling on Getty's passing off claim. She held that it added nothing to her trade mark findings and that certain points had not been adequately argued. The claim was effectively dismissed in light of the trade mark conclusions.[^17][^24]
Following the main judgment, Mrs Justice Smith held a consequentials hearing on 16 and 17 December 2025 dealing with costs, the form of order and permission to appeal. Getty was ordered to pay 69.4 percent of Stability AI's costs on the standard basis, with an interim payment on account of approximately £4,395,656.89. The court ordered a damages inquiry in relation to Stability AI's limited acts of trade mark infringement. Each party was reported to have spent more than £7 million on the litigation by the time of the consequentials hearing.[^10][^25]
Justice Smith granted Getty permission to appeal the dismissal of its secondary copyright infringement claim, holding that the proposed appeal "does, in my judgment, have a real prospect of success" and concerns "a pure question of law, namely a matter of statutory construction on which the minds of reasonable lawyers may differ." Stability AI was refused permission to appeal the limited findings of trade mark infringement, although it retained the option of seeking permission directly from the Court of Appeal. Getty had until 3 February 2026 to lodge its appeal.[^10][^25]
Getty Images issued a public statement after the judgment characterising the trade mark finding as a significant intellectual property victory and noting in particular that the court had held that an intangible AI model could in principle face secondary copyright liability on the same conceptual footing as a physical good. The statement said Getty would seek to "leverage the factual findings from this UK decision in its ongoing U.S. case against Stability AI" and urged the UK government to introduce stronger transparency requirements for AI training data to reduce the cost of enforcement for rights holders.[^26]
Industry counsel were broadly aligned in describing the judgment as a significant practical win for AI developers in the UK, even while emphasising the narrow grounds of decision. Law firms including Bird & Bird (which acted for Stability AI in the litigation), Osborne Clarke, Mayer Brown, DLA Piper and Latham & Watkins published analyses highlighting that the decision did not address the central question of whether unlicensed scraping of copyright works for the purpose of AI training is itself lawful under UK law, because Getty had abandoned its primary copyright claim before that issue could be decided.[^4][^5][^9][^23][^27]
Commentary from academic commentators, including Andres Guadamuz, reader in intellectual property law at the University of Sussex, characterised the judgment as a landmark first English decision on AI and copyright while emphasising that the substantive territorial limits of UK copyright law in relation to AI training remained unresolved by the ruling.[^28]
As of May 2026, the case in the Northern District of California remains pending. Following the August 2025 refiling, the action has been proceeding before Magistrate Judge Lisa J. Cisneros. No trial date had been set publicly, and the parties had not reached settlement. The refiled complaint maintains the core copyright, DMCA, trade mark and unfair competition claims pleaded in Delaware, and adds a copyright dilution theory.[^7][^8][^22]
The procedural posture of the US case differs materially from the UK case in several respects. US copyright law is centred on a statutory four-factor fair use defence rather than the narrower exceptions in UK copyright law, and US case law on AI training and fair use was, as of May 2026, developing through a series of parallel actions involving other defendants. Getty has indicated publicly that it intends to rely on factual findings from the UK trial, including Justice Smith's findings about the appearance of Getty watermarks in early Stable Diffusion outputs, in the US proceedings.[^26]
The California case is distinct from but related to Andersen et al v. Stability AI Ltd. et al., a separate class-action lawsuit brought by visual artists Sarah Andersen, Kelly McKernan and Karla Ortiz against Stability AI, Midjourney and DeviantArt in the same district (case number 3:23-cv-00201). The Andersen action focuses on the alleged use of artists' works in the LAION training datasets and was scheduled to go to trial on 8 September 2026 as of May 2026.[^29]
Justice Smith's ruling has been characterised by legal commentators as offering important but partial guidance to the AI sector. Three threads of the judgment are likely to be of lasting significance:
First, the holding that an "article" within sections 22 and 23 of the CDPA can be intangible expands the conceptual scope of secondary copyright infringement to include AI models. In principle, a future plaintiff that can demonstrate that model weights or another distributed AI artefact contain reproductions of copyright works might still succeed where Getty did not, because the legal threshold for "article" is now lower than the trial court found prior to the judgment.[^4][^5]
Second, the ruling reinforces, but does not finally decide, the territoriality problem at the heart of UK copyright claims against globally distributed AI training. Because Getty abandoned its primary copyright case on training, Justice Smith did not rule on whether scraping or training conducted abroad but resulting in commercial deployment in the UK could amount to an act of copying within the UK. That question, and the related question of whether UK or EU text-and-data mining exceptions might apply, remains open for future litigation.[^4][^5][^9]
Third, the trade mark findings affirm that an AI developer cannot escape liability simply because users are the immediate prompters of an infringing output, where the developer has trained the model on signs that function as trade marks. The judgment's reasoning has been read as a warning to model developers that training-data choices have downstream trade mark implications that cannot be cured by output-side disclaimers alone.[^23][^27]
The case has reignited UK policy debate about whether the government should introduce a text-and-data mining exception broad enough to cover commercial AI training. Getty's official response to the judgment expressly called on the UK government to introduce greater transparency obligations for AI developers so that rights holders need not incur multi-million-pound litigation costs simply to ascertain whether their works were used. Commentary published on the Kluwer Copyright Blog and elsewhere argued that the November 2025 ruling, far from resolving the question, demonstrated the inadequacy of existing UK statute to address generative AI training and underlined the need for legislative reform.[^26][^30]
The Getty v. Stability AI proceedings illustrate the cross-jurisdictional risk that AI developers face when models are trained on data scraped from globally accessible websites and then distributed in multiple legal systems. Getty's decision to pursue parallel actions in the United Kingdom and the United States meant that Stability AI faced different statutory frameworks for substantially the same factual allegations: UK copyright and trade mark statutes, US Copyright Act and DMCA provisions, and US trade mark and unfair competition law. The UK outcome did not bind the US court and vice versa, but the factual record built in the UK trial, including Getty's evidence about the LAION datasets and watermark generation, is expected to be deployed in the California proceedings.[^26][^31]
Getty's claims sit alongside a broader docket of intellectual property litigation against AI developers. In the United States, parallel actions include Andersen v. Stability AI (visual artists), New York Times v. OpenAI, Bartz v. Anthropic (authors) and Concord Music Group v. Anthropic (music publishers). These cases involve different categories of copyright work, different defendants and different statutory theories, but all turn in part on the relationship between AI training and existing copyright doctrines such as fair use, transformative use and the scope of protectable expression. Getty's UK action is widely seen as the first such case to reach a substantive merits judgment in any common-law jurisdiction.[^9][^29]