Summary
In the case of Getty Images v Stability AI Ltd, the High Court of England and Wales examined whether the Stable Diffusion AI model could give rise to copyright infringement liability. Getty Images argued that Stability AI had used millions of its copyrighted photographs to train the model and that both the AI generated outputs and the model itself infringed copyright. During the proceedings, Getty Images abandoned its claim relating to infringing outputs after the prompts used to generate those outputs were blocked. The court therefore examined only whether the AI model itself constituted an infringing copy and held that Stable Diffusion does not contain or store copies of copyrighted images and therefore cannot qualify as an infringing copy.
Background
Copyright Dispute Between Image Library and AI Developer
Getty Images operates a global licensing business for photographs and visual content created by professional photographers and contributors. Its business depends on licensing millions of images that are used by media organizations, advertisers, publishers, and other commercial users. Stability AI develops generative artificial intelligence systems, including Stable Diffusion, an AI model capable of generating synthetic images based on text prompts entered by users.
Getty Images brought the lawsuit alleging that Stability AI had copied millions of images from Getty Images’ websites without permission and used those images to train different versions of the Stable Diffusion model. According to Getty Images, many of those images were original artistic works in which copyright subsisted and were either owned by Getty Images or exclusively licensed to it. Getty Images initially raised several copyright claims in the case. These included allegations that the training process involved unlawful copying of copyrighted images, that the outputs generated by the AI system reproduced protected works, and that the AI model itself constituted an infringing copy.
During the proceedings, Getty Images acknowledged that there was no evidence that the training and development of Stable Diffusion had taken place in the United Kingdom. It also accepted that the prompts used to generate allegedly infringing outputs had been blocked by Stability AI. As a result, the claims relating to the training process and the allegedly infringing AI outputs were abandoned. The remaining copyright issue before the court concerned whether the Stable Diffusion model itself could be treated as an infringing copy under the secondary infringement provisions of the Copyright, Designs and Patents Act 1988.
Questions Before the Court
1. Whether the Stable Diffusion AI model could be treated as an infringing copy of copyrighted works under the Copyright, Designs and Patents Act 1988.
2. Whether importing, distributing, or possessing the AI model could amount to secondary copyright infringement.
3. Whether the model weights created during the training of the AI model could constitute copies of copyrighted works.
Arguments Presented By the Parties
Getty Images’ Arguments
Getty Images argued that Stable Diffusion had been trained using millions of copyrighted photographs taken from its websites. It contended that the training process resulted in the creation of model parameters that embodied information derived from those copyrighted works. According to Getty Images, these model weights represented a transformed form of the copyrighted images. On this basis, Getty Images argued that the AI model itself should be treated as an infringing copy under the Copyright, Designs and Patents Act.
Stability AI’s Arguments
Stability AI argued that Stable Diffusion does not store copies of the images used during training. It stated that the model consists of numerical parameters that encode statistical relationships learned from training data. According to Stability AI, these parameters cannot be reconstructed into the original images and therefore cannot constitute copies of copyrighted works. Stability AI therefore argued that the AI model cannot qualify as an infringing copy.
Court’s Analysis of Copyright Claims Against the AI Model
The court stated that the secondary infringement provisions of the Copyright, Designs and Patents Act apply to articles that embody or contain infringing copies of copyrighted works.
The court observed that Stable Diffusion is a machine learning model consisting of mathematical parameters created during training. According to the court, the model does not store the training images or contain copies of the copyrighted works used during training.
The court explained that the model parameters influence how the system generates images but do not reproduce or store the original images themselves. In the eyes of the court, the AI model therefore cannot be treated as containing copies of the copyrighted works.
The court also noted that Getty Images did not contend that Stable Diffusion itself stored copies of the copyrighted images. Instead, Getty Images argued that the creation of model weights should itself be treated as an infringing copy. The court rejected this argument and held that the model weights do not constitute copies of the copyrighted works.
Findings
The court held that Stable Diffusion does not contain copies of Getty Images’ copyrighted photographs.
It therefore concluded that the AI model cannot qualify as an infringing copy under the Copyright, Designs and Patents Act 1988.
As a result, the claim that Stable Diffusion itself constituted an infringing copy failed.
Case Citation
Getty Images US Inc v Stability AI Ltd, 2025 EWHC 2863 Ch.
Disclaimer
This case blog is based on the author’s understanding of the judgment. Understandings and opinions of others may differ. An AI application was used to generate parts of this case blog. Views are personal.