Introduction
Getty Images has initiated a legal battle against Stability AI [4], a company specializing in open-source generative artificial intelligence (GenAI). The lawsuit centers on allegations of unauthorized use of Getty’s copyrighted materials to train Stability AI’s model, “Stable Diffusion.” This case is poised to become a landmark in the realm of intellectual property (IP) law, particularly concerning the use of unlicensed data for AI training.
Description
Getty Images has initiated legal proceedings against Stability AI [1] [4], an open-source generative artificial intelligence (GenAI) company [1], alleging unauthorized use of millions of its copyrighted images [4], videos [4], and illustrations to train the AI model, ‘Stable Diffusion.’ The claims include copyright infringement [3], database right infringement [3] [4] [5], trademark infringement [3] [4], and passing off [3] [4]. Getty asserts that over half of the works in question are original and were unlawfully scraped from its websites without permission. The case serves as a pivotal test for intellectual property (IP) infringement claims related to the use of unlicensed training data by AI developers [5], particularly as Stability AI is accused of reproducing substantial parts of Getty’s copyrighted works and displaying its protected watermarks.
In the ongoing case [1] [4], the claimants [1] [2] [3] [5] [6], which include Getty Images (US) Inc and several companies controlled by Getty Images Holdings Inc., have sought to have Thomas M Barwick Inc act as a representative for a class of approximately 50,000 individual copyright owners under CPR 19.8 [1]. Barwick asserted that all legal rights to the works were assigned to Getty and that these works were infringed by Stability AI [6]. However, during a recent case management conference, Stability AI contested the sixth claimant’s ability to represent the class [1], arguing that the varying agreements and levels of exclusivity among individual copyright owners required separate assessments [3]. The High Court [2] [3] [5] [6], presided over by Mrs [1]. Justice Smith [1], evaluated whether the class was adequately defined and whether its members could be identified [1]. The judge expressed concerns about the practicality of identifying each infringing image [4], noting the disproportionate resources required for such an endeavor [4]. The claimants attempted to refine the class definition to include only those who had exclusive licenses with Getty and whose works were used in training Stable Diffusion [6]. However, the absence of a definitive list of the works used for training complicated the identification of class members [6].
Ultimately, the court determined that the class definition relied on the resolution of the central issue of copyright infringement [1], which is to be decided at trial [1], thus complicating the representative action’s viability [1] [5]. Consequently [4], permission for the representative claim was denied [3] [4]. The judge concluded that the determination of which works were used in training Stable Diffusion could not be established at this time [6], preventing the identification of the proposed class [6]. Even if the judge’s interpretation of the claim’s scope under CPR 19.8 was incorrect [6], she was not convinced that the representative claim should proceed to trial [6]. The claimants also sought to pursue their claims without including copyright owners with concurrent rights [6], but this was rejected due to insufficient evidence and a lack of a properly pleaded application [6].
Stability AI also sought to dismiss claims of training and secondary infringement [2], arguing a lack of evidence for UK-based activity and that UK law applies only to tangible items [2]. However, the High Court rejected this dismissal [2], allowing Getty’s claims to proceed [2], including an amended claim regarding the model’s “image-to-image” feature [2]. The judge noted that the absence of a definitive list of copyrighted works complicated the identification of class members [2], but indicated that the claimants could reapply [2], leaving the possibility of a representative action open for exploration during the upcoming trial [2].
The first trial to establish liability is set to commence on June 9, 2025 [4], and the case is expected to have significant implications for intellectual property law in relation to generative AI [4], raising essential questions about the rights of creators and the data needs for AI training [4]. The ongoing developments in the Getty v [2]. Stability AI case coincide with proposed UK copyright reforms aimed at requiring AI developers to disclose the content used in training their models. Such transparency could facilitate rights holders in verifying unauthorized use of their works and streamline the formation of representative actions [2], ultimately influencing the balance between fostering AI innovation and protecting intellectual property rights [2].
Conclusion
The Getty Images lawsuit against Stability AI is set to be a defining case in the intersection of AI development and intellectual property law. As the trial progresses, it will likely shape future legal standards regarding the use of copyrighted materials in AI training. The outcome could influence ongoing discussions about copyright reforms, potentially leading to greater transparency and protection for rights holders while balancing the need for innovation in AI technologies.
References
[1] https://www.penningtonslaw.com/news-publications/latest-news/2025/court-breaks-up-representative-action-romance-getty-images-v-stability-ai
[2] https://dwfgroup.com/en/news-and-insights/insights/2025/2/pixelated-predicaments
[3] https://www.lexology.com/library/detail.aspx?g=3c702601-4d05-4f57-921d-066f665c7925
[4] https://www.lexology.com/library/detail.aspx?g=3594c721-eedb-465c-9ddc-5d4612a486ce
[5] https://www.lexology.com/library/detail.aspx?g=a6865270-1f88-45c1-b890-380b40bbb2a7
[6] https://www.wiggin.co.uk/insight/group-litigation-court-rejects-representative-action-in-getty-and-stability-ai-dispute/
