Facts
The Getty Images companies sued Stability AI, alleging that it had scraped millions of images from Getty's websites without consent to train and develop Stable Diffusion. During the proceedings Getty accepted that there was no evidence that training took place in the United Kingdom and abandoned its training claim. It also abandoned its claim over infringing outputs after the prompts concerned were blocked, and it dropped its database right claim. What remained were a claim that importing and making the model available in the UK was secondary infringement, trade mark claims based on Getty and iStock watermarks appearing in some outputs, and passing off.
Question
The main question was whether an AI model trained abroad could be an 'article' that is an 'infringing copy' under sections 22 and 23 of the Copyright, Designs and Patents Act 1988 (CDPA). Getty argued that the model's making would have infringed copyright had it taken place in the UK. The court also had to decide whether watermarks in outputs infringed Getty's marks under sections 10(1), 10(2) and 10(3) of the Trade Marks Act 1994.
Decision
Joanna Smith J accepted that an 'article' under the CDPA may be intangible, but held that a model such as Stable Diffusion, which does not store or reproduce any copyright works and has never done so, is not an 'infringing copy'; the secondary infringement claim was dismissed. The trade mark claim succeeded in part. iStock watermarks generated by v1.x models accessed through DreamStudio or the Developer Platform, and Getty watermarks generated by v2.x models, infringed, but only on the basis of specific examples. There was no infringement for SD XL and v1.6, the section 10(3) claim failed, passing off was not decided and additional damages were refused.
Why it matters
The first comprehensive UK judgment on generative AI and copyright did not examine copying during training, because of where the training took place. Its core is whether a trained model can be a copy of the works it learned from, an approach that differs from the Munich decision in GEMA v OpenAI, which treated memorisation as reproduction. The watermark findings show that marks appearing in outputs can expose a model provider to trade mark liability.
Related stages
In December 2025 Joanna Smith J granted Getty permission to appeal to the Court of Appeal on secondary infringement and refused Stability AI permission to appeal the trade mark findings.