IL-2023-000007 - [2025] EWHC 2863 (Ch)

REMAINING OUTSTANDING ISSUES

One of the outstanding issues for trial is the number of Visual Assets and Copyright Works used in training. However, there is a dispute between the parties as to whether this is an issue that should still be determined, notwithstanding that Getty Images have abandoned the Database Rights Claim, the Training and Development Claim and the Outputs Claim.

Getty Images submit that the numbers are relevant to at least three issues: (i) secondary infringement of copyright “because Stability’s admissions do not cover v1.1-1.3”; (ii) trade mark infringement and passing off “because the likelihood of a watermark appearing in an output is dependent on the frequency with which the watermark appears in the training data”; and (iii) additional damages “because when the court comes to consider the issue of flagrancy, a relevant factor to weigh in the balance will be the scale of any infringement”.

Stability accepted at the PTR that “many” Copyright Works were used to develop and train Stable Diffusion. In its statements of case it has admitted that “at least some” Copyright Works were used in training v2.0, that at least some images from the Getty Images Websites were used during training; and that one or more of Sample Works A1-A17 were used in the training of v1.4, v2.x, all versions of SD XL and v1.6. It has also admitted for the purposes of these proceedings that there are approximately 12.3 million Visual Assets in LAION 2B-en. It contends that there is no reason for the court to try to make any specific findings going beyond these admissions as to either Visual Assets or Copyright Works. I am inclined to agree. As to the former, it is in any event unclear why a finding as to Visual Assets could continue to be of relevance given the abandonment of the Database Rights Claim. As to the latter, Stability correctly points out that any attempt to make findings on the number of Copyright Works would raise complex issues - not least because Getty Images’ case requires them to satisfy the court that the assets used in training were images whose copyright was owned by one of the Claimant companies or whose copyright has been licensed to one of the Claimant companies under one of the Annex 3 Licence Agreements. Insofar as Getty Images is inviting the court to determine the rights ownership position for each work, that is a fact sensitive question for each work and plainly impractical.

I decline to make any finding as to the number of Visual Assets or Copyright Works used in training. In addition to the points made by Stability and set out above, and taking each of Getty Images’ reasons for inviting me to do so in turn:

Finally, and in addition to the fact that I cannot see the need to make any finding as to the number of Visual Assets or Copyright Works in the training data, I do not consider that the available evidence would enable me to do so. I note, in particular that I could attach little weight to Ms Cameron’s fourth witness statement owing to the way in which that evidence was produced and Ms Cameron’s limited understanding of the steps taken to put it together. In their written submissions, Getty Images put their case on numbers no higher than that “it cannot sensibly be suggested that the development and training of each of the sub-versions of Stable Diffusion v1.x, v2.x and XL involved substantially fewer than millions of Visual Assets and Copyright Works”. I can see no reason to make such a vague (and ultimately irrelevant) finding in the circumstances of this case. Given the way in which the case has been advanced, I cannot see that it would assist me in arriving at any reliable conclusion as to the scale of watermark* generation on synthetic images by users in the UK.

For all the reasons I have given, I decline to make any such finding.

There is no dispute between the parties as to the relevant legal principles. Getty Images referred me to the Intellectual Property (Enforcement, etc.) Regulations 2006 which implement Directive 2004/48/EC and to section 97(2) CDPA (which is concerned only with copyright infringement).

Getty Images’ pleaded case in support of additional damages relies upon three matters: first that Stability knows or has reason to believe that Stable Diffusion is an infringing copy of the Copyright Works; second that Stability has acted with a “cavalier attitude” to the rights of the Claimants, “infringing numerous different intellectual property rights on a blatant and widespread scale”; and third that Stability has relied upon the infringement of copyright and database right for the successful operation of its business and Stable Diffusion has been, or has the potential to be, used by end users who would otherwise have used the Getty Images Websites.

Given that I have dismissed the Secondary Infringement Claim and that both the Outputs Claim and the Training and Development Claims were abandoned at trial, I cannot see how either the first or third of the particulars pleaded by Getty Images goes anywhere. As Stability submits, the position is that the Models are producing novel content in the United Kingdom against which Getty Images have no claim. Where the court cannot conclude that the training was an infringement (because it took place in the United States, or elsewhere) and has found that there have been no acts of secondary infringement, there can be no basis for additional damages for infringement of copyright under section 97(2) CDPA. I dismiss the claim for additional damages under that section.

As for the second of Getty Images’ particulars, it cannot possibly be maintained in light of my findings that Stability has infringed “numerous different intellectual property rights on a blatant and widespread scale”. While I have found instances of trademark infringement, I have been unable to determine that these were widespread, or that they continued beyond the release of v2.x. In the circumstances, there is, in my judgment, no basis whatsoever for a claim for additional damages under the broader provisions of the 2006 Regulations. Getty Images may be able to maintain such a case in the jurisdiction where the Model was in fact trained, but there is no basis for that case in this jurisdiction.