[2025] EWHC 2863 (Ch) Case No: IL-2023-000007
Overview: Getty Images v Stability AI Decision
The UK High Court rendered a historic ruling today in the Getty Images v. Stability AI case. The legal, technology and creative sectors both keenly followed this case, which brought up important issues regarding the application of copyright and trademark rules to AI models that have been trained on enormous volumes of online visual content. The ruling by Mrs Justice Joanna Smith DBE is the first of its kind in the UK and provides crucial guidance on striking a balance between AI advancement and intellectual rights.
The Case’s Background
Getty Images, a world leader in licensed photography and visual content, claimed that Stability AI had trained its generative AI model, Stable Diffusion, using millions of its copyrighted photos without permission. To acquire patterns for creating new images from word prompts, the model—a kind of latent diffusion system—was trained on billions of photographs from the internet. In 2022, Stability AI made it available as open-source software, allowing anybody to freely produce images.
Getty filed a lawsuit with the High Court in January 2023, alleging that Stability AI had violated its intellectual property rights by unauthorizedly scraping Getty’s watermarked photos. The conflict became emblematic of the larger conflict between tech companies warning that stringent laws could impede the development of AI and creative industries seeking protection.
When the matter went to trial in 2025, Getty dropped many of her initial allegations. Since Getty was unable to provide evidence that the training took place in the UK, the main copyright claim pertaining to the training procedure was dropped. After Stability AI reduced problematic prompts and Getty got most of the help it asked for, other claims—like those about possible copyright violations from some AI-generated outputs and database rights—were dropped.
In the end, the High Court had to make decisions on two main matters: (1) whether Stability AI was responsible for passing off and trademark infringement because AI-generated photographs had Getty’s watermark, and (2) if Stability AI was liable for secondary copyright infringement because of the Stable Diffusion model’s propagation.
Getty’s Allegations of Copyright Violation
Getty claimed that the Stable Diffusion model was an “infringing replica” of its photos, having been made using illegal copies of those images. Getty contended that releasing the trained model in the UK amounted to secondary infringement, analogous to selling or importing infringing goods, since direct infringement claims were no longer current. Additionally, Getty asserted that Stability AI’s model illegally copied its trademarks. According to Getty, some AI-generated photos featured warped versions of the “Getty Images” or “iStock” watermarks, which might give the impression that the outputs were approved or licensed by the business.
Passing Off
Getty claimed that their watermark’s inclusion on AI-generated photos amounted to deception under common law, damaging its brand and goodwill.
The Court’s Judgement
Whether the Stable Diffusion model qualified as an “infringing copy” was the main query. Secondary infringement occurs when someone deals with an unauthorised copy of a work—by importing, selling, or distributing it—knowing it was made illegally, according to UK copyright law.
The judge acknowledged that a digital file might legally be considered an “article” but concluded that Stable Diffusion lacked or replicated any of Getty’s real photos. Rather than keeping the actual images, it recorded statistical representations of image patterns. The court compared this to a human artist who learns by doing without keeping the originals. The model cannot be considered an infringement copy since it does not represent Getty’s works.
As a result, the court rejected Getty’s secondary infringement claim, finding that copyright was not infringed by the model’s distribution or use in the UK. The issues of trademarks and passing off were demonstrated by evidence that some AI-generated photos contained subtle emulations of iStock or Getty watermarks. The court determined that there was limited trademark infringement in these cases.
When the watermark on AI-generated photographs was identical, the court determined that there had been infringement of the iStock mark, which amounted to the unapproved use of a protected sign on identical goods.
The court found that certain generated watermarks for the Getty Images mark met the requirements for infringement because they were sufficiently similar to confuse consumers.
There was no evidence that Stability AI purposefully took advantage of or harmed Getty’s brand; thus, the court rejected more general allegations of dilution or unfair advantage.
Since the trademark analysis had addressed the same facts, the passing-off claim was not considered separately.
The court stressed that these conclusions had a limited reach because there was no proof that more recent iterations of Stable Diffusion were still reproducing Getty or iStock watermarks, and only a few isolated instances were proven.
Result
Getty won a narrow victory. In a few cases, it was successful in establishing trademark infringement, but its main copyright claim was completely unsuccessful. The court acknowledged Stability AI’s efforts to reduce the possibility of abuse and pointed out that the confirmed instances of watermark replication were small and historical. It’s likely that Getty will only receive minor remedies, such as injunctive relief or nominal damages. According to the ruling, which generally favoured Stability AI, a trained AI model that does not store original copyright information is legal under current UK legislation.
Implications for AI Training and Copyright in the Creative Industries
The decision makes it clear that using copyrighted material to train an AI model and then sharing the model does not always infringe UK copyright law, especially if the training takes place outside of the country and the model has no stored copies. This gives AI developers peace of mind, but it also reveals a legal loophole for creators. Many in the creative industry argue that handling the sources and uses of AI training data requires new laws. Understanding what an AI model is essential to comprehending the logic.
The model learns what a cat or dog looks like by seeing images of them connected to words like “cat” and “dog”. Without referring to the original photographs, it can now faithfully replicate what a picture of a dog or cat should seem like.
Compliance with Trademarks and Output Controls
The ruling cautions AI firms that even if their models inadvertently replicate brand identities, they could still be held liable under trademark law. By using quick filters and better model training, developers can avoid outputs that contain protected marks or trademarks.
Outlook for Policy and Legislation
Since then, Getty Images and other interested parties have demanded more openness about AI training datasets and legislative measures to guarantee equitable remuneration for content producers. Legislators are looking into potential changes, such as establishing opt-in and opt-out procedures for rightsholders or elucidating the exceptions for text and data mining. The UK government is still seeking advice on how to strike the correct balance between creators’ rights and innovation.
Global Background
This ruling contributes to an expanding international discussion. In the US, where “fair use” issues will predominate, Getty is pursuing rival litigation. By making a distinction between training on copyrighted works and the actual use or dissemination of AI models, the UK verdict could have an impact on future cases globally.
In conclusion
An important turning point in the relationship between AI and intellectual property law was reached by the High Court in its ruling in Getty Images v. Stability AI. It affirms that current trademark protections still apply to AI-generated content and states that a generative AI model trained on copyrighted information is not always an infringing copy. The ruling provides an early legal roadmap for how AI and intellectual property can coexist in the years to come by carefully balancing the protection of creative rights with the advancement of technology.
