The High Court has recently handed down judgment on Stability AI’s application for reverse summary judgment and strike out in the high-profile Getty Images v Stability AI litigation. Earlier this year, Getty sued Stability for copyright, database right and trade mark infringement, as well as passing off, in connection with the training, development and deployment of the Stable Diffusion AI model.
In its recent application, Stability sought unsuccessfully to defeat some of Getty’s claims on a summary basis, namely that:
- Training and development of Stable Diffusion (as opposed to merely answering prompts) involved copying Getty’s copyright works (and extracting them from Getty’s database) in the UK and therefore infringed Getty’s rights;
- Stability “imported Stable Diffusion” into the UK and so dealt in an “infringing copy” for the purposes of copyright; and
- Stability infringed and passed off Getty’s marks in certain of Stable Diffusion’s output. This argument was dropped at the hearing, so will be determined at trial.
Summary judgment was not sought in respect of all of Getty’s claims: (i) an allegation of copyright infringement for authorisation of and communication to the public where Stable Diffusion is used to generate an image that reproduces the whole or a substantial part of copyright works; and (ii) database right infringement by re-utilisation of a substantial part of the database – in both cases by users in the UK.
The bar for a successful summary judgment application is high. Stability needed to show that Getty’s claims had no real prospect of success and that there were no compelling reasons why the claims should be disposed of at trial. Case law has established that the court should not conduct a mini-trial and that a real prospect of success requires a claimant’s claim to be more than merely arguable (rather than fanciful).
Training, development and territoriality
The first basis for Stability’s application: copyright and database rights are territorial rights and only copies made (or extraction carried out) in the UK can infringe. Stability argued that there is no prospect that Getty could evidence that the training or development of Stable Diffusion took place in the UK.
In support of its argument, Stability filed evidence to the effect that all its computations happened outside the UK, and none of its UK staff were involved in or had the right equipment to engage with training. This evidence, however, was held to be inconclusive and public statements from Stability appeared to contradict its evidence before the court. For example, YouTube and podcast interviews showed that Stability had hired AI programmers from both Ukraine and Russia and “fast tracked their residence in the UK”, and there was at least a chance that these workers had been assigned to Stable Diffusion.
Overall, Smith J concluded that it was more appropriate for the claim to be dealt with at trial, when the full and precise nature of Stable Diffusion’s development can be identified.
The standard for summary judgment is demanding, and Stability’s failure to achieve it in this instance does not affect its overall merits.
Secondary infringement through importing, possessing or dealing in an “infringing article”
Secondly, the Copyright, Designs and Patents Act 1988 (CDPA) restrains importing, possessing or dealing in an article in the UK which is an “infringing copy”. An “article” is an “infringing copy” if making it in the UK (hypothetically) would have constituted an infringement of copyright in the UK. But can an intangible thing such as an AI model hosted on a cloud server be an “article”?
Stability again asked the court to strike out Getty’s plea of infringement by importation into the UK on the basis that “articles” in the CDPA is limited to tangible things, and so did not encompass dealings in intangibles (such as making available software on a website). The judge concluded that this issue needed full argument and so was unsuitable for summary judgment.
Getty’s amendment application
For its part, Getty asked for permission to amend its pleading to include a new image-to-image feature of Stable Diffusion. The feature allows a user to upload an image and determine how closely they wish the AI-generated output to align to the source image using an “image strength” slider. Stability argued that in so far as there is any copying it is brought about by the user: they will be the one uploading the source image (e.g. an unlicensed Getty Images copyright work) and specifying the level of similarity required. Smith J held, however, that Getty’s claim did have a real prospect of success, based on uncertainty regarding the technical involvement of Stability in producing the AI-generated output and given its similarity with Getty’s text-to-image claim. Permission to amend was therefore granted.
Social Media cookies collect information about you sharing information from our website via social media tools, or analytics to understand your browsing between social media tools or our Social Media campaigns and our own websites. We do this to optimise the mix of channels to provide you with our content. Details concerning the tools in use are in our privacy policy.