- Initiation of the Case: Getty Images is challenging Stability AI in the UK High Court over alleged misuse of copyrighted images for their image-generating tool, Stable Diffusion.
- Core Legal Disputes: The case centers around issues of copyright infringement, trademark violations, and database rights, with Getty claiming that Stability AI improperly used its copyrighted works to train the Stable Diffusion model.
- Implications for AI Development: The outcome of this case could redefine the legal landscape for AI development in the UK, impacting how generative AI models can legally use publicly available content for training purposes.
One of the most significant legal battles in recent times kicked off last month in the UK High Court: Getty Images sued Stability AI, the company behind the image-generating tool Stable Diffusion. This case might just redefine how artificial intelligence can be developed, trained and deployed in the UK. Whether you are a tech specialist, a lawyer or someone who has typed "make me an image of a dog in black tie" into an AI image generator, this one matters.
The Brief
Getty Images owns one of the world's largest libraries of photographs and videos. Many of its images carry the well-known Getty watermark and are stored in a highly organized system that the company claims has cost hundreds of millions of dollars to build and maintain.
Enter Stable Diffusion, a tool developed by Stability AI that lets users generate realistic synthetic images from text and image prompts. Think of it as visual alchemy – the user types "Medieval robot wedding" and the result is something that's often eerily plausible. The controversy? The model was trained on numerous images scraped from across the web, millions of which, Getty claims, came from their archive of copyright protected works.
In response, Getty sued Stability in early 2023, alleging copyright infringement, trade mark infringement, passing off, and database right infringement. No pressure then.
The Core Legal Issues
The issue of copyright infringement lies at the heart of this case. Getty originally brought three main copyright claims. The first related to the inputs used to train Stable Diffusion, which Getty argued were illegally fed into the model given their status as copyrighted works. The legality of the outputs was also disputed, given their derivation from copyrighted material. Thirdly, the legal status of Stable Diffusion itself was brought into doubt, with Getty contending that it was, in fact, an illegal product containing copyrighted (and therefore, protected) material.
Stability AI countered that training took place outside the UK, so UK copyright law doesn't apply. As for the outputs, the company argued that the users — not Stability — are the ones generating images, and therefore Stability is not responsible. Stability also invoked a little-used UK legal defence: fair dealing for "pastiche", which allows copying for parody or stylistic imitation.
Most interestingly, Getty pushed a novel legal argument: that the model itself is an infringing "article" under UK law — a claim Stability dismisses saying their model is just intangible information.
But here is the twist: Getty dropped the input and output claims towards the end of the trial, so the only copyright issue still alive is whether Stable Diffusion itself is an 'article' which is an 'infringing copy'.
Trademarks
Getty also claims that some output images created by Stable Diffusion contain its watermark, which could mislead viewers into thinking those images are genuine Getty content. This is classic trademark territory.
Stability responded that the examples Getty produced were contrived and unlikely to occur in real-world use. They also argue that even if such images did appear, they weren't being sold or used "in the course of trade" — meaning, ultimately, that there's no trademark infringement.
Database right
Under UK law, companies can claim a unique intellectual property right over databases they've invested in — if the maker of the database has made a 'substantial investment' in obtaining, verifying or presenting its contents. Getty claimed that Stability 'extracted' a substantial part of its database and infringed its database right.
Stability pushed back, saying the training of Stable Diffusion was conducted outside the UK. This issue is now academic as Getty also dropped the database claim near the end of the trial.
Why Does This All Matter?
The decision in this case could reshape the rules for how generative AI is developed in the UK. If the court sides with Getty, even partially, it could make it significantly riskier to train AI models on freely available internet content if certain acts take place in the UK. Conversely, a ruling in Stability's favour might greenlight a free-for-all on scraping and training at the expense of authors and copyright owners in the UK. The British government is currently considering reforming copyright law to make it easier to develop AI models in the UK, whilst also protecting the rights of copyright owners.
This isn't just a UK issue. Getty has also sued Stability in the US, where the legal framework — and the fair use doctrine — could lead to a very different outcome. But with the UK's High Court now grappling with how to classify a complex AI system like Stable Diffusion under old-school IP laws, this judgment could become a benchmark for how other countries approach the same challenge.
Originally Published by Tech Monitor
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
