Getty Images (UK) initiated legal proceedings against Stability AI in the High Court of Justice of England and Wales in January 2023, alleging that the company's Stable Diffusion model was trained on millions of copyrighted photographs without authorization. The case, formally styled Getty Images (UK) Ltd and others v Stability AI Ltd, became one of the first major UK lawsuits to test how copyright law applies to Generative AI systems. Getty sought damages and an injunction to prevent further use of its images in training data, arguing that Stability AI's scraping of its website and databases infringed both copyright and database rights under UK law.
The dispute arose from Stability AI's development of Stable Diffusion, a Deep learning model that generates images from text prompts. Stability AI, founded in 2020 and headquartered in London, released Stable Diffusion in August 2022. The model was trained on a large dataset known as LAION-5B, which contained billions of image-text pairs scraped from the internet, including many images hosted on Getty's platforms. Getty contended that this scraping and subsequent use constituted unauthorized reproduction and communication to the public of its protected works. Stability AI argued that the training process involved transformative use and that the model did not store or reproduce individual images directly, but rather learned statistical patterns.
The High Court case proceeded through procedural stages in 2023 and 2024. In July 2023, the court rejected Stability AI's application to strike out parts of Getty's claim, allowing the case to proceed to a full trial. The court found that Getty had raised arguable claims regarding copyright infringement, particularly concerning the reproduction of images during training and the potential for the model to output near-identical copies of source images. A case management conference in early 2024 set a trial date for late 2025, with both sides engaging in extensive disclosure of technical documents and training data logs.
Legal Claims and Defenses
Getty's primary claim rested on the reproduction right under the Copyright, Designs and Patents Act 1988. The company argued that Stability AI's training process necessarily involved copying images into the model's parameters, even if the copies were not directly accessible. Getty also invoked the UK's database right, which protects substantial investment in compiling data. Stability AI countered that the LAION dataset was publicly available and that the training process was analogous to a human viewing images to learn artistic styles. The company further argued that any infringement was transient and incidental, falling under exceptions for temporary reproductions.
The court's July 2023 ruling, delivered by Mr Justice Birss, addressed several preliminary issues. Birss held that Getty's claims were not obviously doomed to fail, noting that the question of whether training constitutes infringement was a matter for full trial. He also allowed Getty to amend its claim to include allegations about the model's ability to generate images that closely resemble specific photographs, a point that could strengthen the infringement case. Stability AI's defense that the model was trained on data lawfully obtained from third parties was deemed insufficient to dispose of the claim at an early stage.
Technical Aspects of the Case
The case required the court to understand how Neural network models like Stable Diffusion operate. Stability AI's technical experts explained that training involves feeding millions of images through a U-Net architecture, which learns to denoise random noise into coherent images. The model stores weights - numerical values - that encode statistical relationships, not pixel-by-pixel copies. However, Getty's experts argued that the model can memorize and reproduce specific training images, especially when prompted with certain text. This memorization phenomenon, documented in academic research on Large language models and image generators, became a central factual dispute.
The LAION-5B dataset, created by the non-profit LAION organization, was assembled by crawling web pages and extracting image URLs and alt-text. Stability AI used a filtered subset of this dataset to train Stable Diffusion. Getty's images appeared in the dataset because many websites embedded Getty photos without authorization. The case raised questions about whether Stability AI had a duty to verify the copyright status of images in the dataset, or whether it could rely on the dataset's public availability. The court heard evidence about the scale of the scraping, with Getty estimating that over 12 million of its images were included in the training data.
Procedural History and Key Rulings
After the July 2023 strike-out application failed, the case moved to disclosure. In November 2023, the court ordered Stability AI to provide detailed logs of its training data, including the specific images used and the dates of scraping. Stability AI initially resisted, citing commercial confidentiality, but the court ruled that transparency was necessary for Getty to assess its claims. In February 2024, a further hearing addressed the admissibility of expert evidence on AI model behavior. The court allowed both sides to submit expert reports from computer scientists, including specialists in Machine learning and copyright law.
A significant development occurred in April 2024 when the court granted Getty permission to add a claim for secondary infringement against Stability AI's users. This claim alleged that individuals who used Stable Diffusion to generate images infringed copyright by creating derivative works. Stability AI argued that its users were not liable because the model did not reproduce protected expression. The court, however, held that this was a triable issue, particularly given evidence that some users had generated images closely resembling famous photographs. The case was consolidated with a related claim against Stability AI's parent company, Stability AI Ltd, which was incorporated in the UK.
Industry and Policy Context
The lawsuit unfolded against a backdrop of global litigation over AI training data. In the United States, Getty filed a separate suit against Stability AI in the District of Delaware in February 2023, which was later transferred to the Northern District of California. The UK case was seen as a test for European copyright law, as the UK had not yet adopted the EU's 2019 Directive on Copyright in the Digital Single Market, which included a text-and-data-mining exception. The UK government had considered introducing a similar exception but shelved plans in 2023 amid pressure from creative industries. The outcome of the Getty case could influence whether the UK adopts a more permissive or restrictive approach to AI training.
The case also drew attention to the broader ecosystem of Generative AI companies. Stability AI was one of several firms, including OpenAI and Anthropic, that faced lawsuits over training data. In the UK, the case was closely watched by publishers, photographers, and technology companies. The Artificial intelligence industry argued that overly strict copyright rules would stifle innovation, while creative professionals contended that AI companies were profiting from unlicensed use of their work. The court's eventual decision could set a precedent for how UK courts interpret copyright in the context of AI, potentially affecting future cases involving Google DeepMind or other UK-based AI developers.
Expert Witnesses and Technical Evidence
The trial, scheduled for late 2025, was expected to feature testimony from leading computer scientists. Getty planned to call experts in Computer vision and Deep learning to demonstrate how Stable Diffusion's architecture could memorize training images. Stability AI intended to rely on experts who would argue that the model's outputs are novel creations, not copies. The court also received evidence on the effectiveness of Data Augmentation techniques and Model Pruning in reducing memorization. Both sides commissioned studies to test whether specific Getty images could be reproduced by prompting the model with their alt-text descriptions.
One key technical question was whether the training process involved 'communication to the public' under UK law. Getty argued that by making the model available online, Stability AI communicated the embedded copies of its images to users. Stability AI responded that the model does not transmit images directly, but rather generates new images based on learned patterns. The court's treatment of this issue could have implications for how Transformer (architecture)-based models are treated under copyright law. The case also touched on the role of Positional Encoding and Multi-Head Attention in the model's ability to associate text with visual features, though these technical details were secondary to the core legal questions.
Potential Outcomes and Implications
If Getty prevails, the court could issue an injunction requiring Stability AI to remove all Getty images from its training data, a technically challenging task given that the model's weights do not contain identifiable copies. Alternatively, the court could award damages based on the number of infringing images used, which Getty estimated at over 12 million. Stability AI, which faced financial difficulties in 2023 and 2024, might struggle to pay substantial damages. A ruling in favor of Stability AI would clarify that training on publicly available data does not infringe copyright, potentially encouraging more AI development in the UK.
The case's outcome could also affect the OpenPanel debate on AI regulation. The UK government had initially adopted a pro-innovation stance, but the lawsuit highlighted the need for clearer rules. In 2024, the UK's Intellectual Property Office issued non-binding guidance on AI and copyright, but the High Court's decision would carry more weight. The case was also watched by international bodies, including the World Intellectual Property Organization, which had been discussing AI and copyright since 2020. As of early 2025, the case remained pending, with both sides preparing for the trial. The court's judgment, whenever it comes, would likely be appealed, meaning the final resolution could take years.
Related Proceedings and Global Impact
The UK case was one of several legal actions against Stability AI. In addition to the US suit, the company faced claims in Germany and Japan. The UK proceedings were considered particularly significant because the UK is a major hub for AI research, with institutions like University of Oxford and University of Toronto contributing to the field. The case also intersected with debates about OpenAI's and Anthropic's use of training data, though those companies were not parties to the UK suit. The court's reasoning on technical issues, such as whether model weights constitute copies, could influence how other jurisdictions address similar questions.
The case highlighted the tension between Generative AI innovation and intellectual property rights. Stability AI argued that its model was a tool for creativity, enabling users to generate new artworks. Getty countered that the tool was built on the unauthorized labor of photographers. The High Court's role was to interpret existing law, not to set policy, but its decision would inevitably shape the regulatory landscape. As the trial approached, both sides continued to refine their arguments, with legal costs estimated to exceed £10 million. The case stood as a landmark in the evolving relationship between Artificial intelligence and copyright law in the United Kingdom.