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

Turning then to the Getty Images’ Marks and Signs, I need to deal with the two Signs separately.

Stability contends that because the watermark* on the Obama Image was generated in the course of the Getty Watermark Experiments, I should disregard it as unrealistic and unrepresentative of a watermark* that would be produced in real life. For all the reasons I have already given, I agree. To my mind there is no evidence that such a clear Getty Images watermark* has been produced by any of the Models in issue using a real world prompt and accordingly I do not consider that it is appropriate to use this Sign for the purposes of the comparison.

However, if I am wrong about this, then I agree with Getty Images that the Sign on the Obama Image has clear identity, aurally, visually and conceptually with the Getty Images word mark. There are only two differences between Mark and Sign:

I arrive at a similar conclusion in relation to the comparison between the figurative Mark and Sign. Here the only difference is the apostrophe and, as I have said, that would go unnoticed by the average consumer.

I did not understand Stability seriously to dispute this analysis. It points out that the photographer’s name under the Sign is garbled and that (having regard to the evidence of Mr Stanley) the average consumer would expect the photographer’s name to be legible. It also contends that because the garbled name is co-located with the other wording within the boundary of the watermark*, the average consumer will perceive it as part of the Sign and will not therefore regard the Sign as identical to the Marks.

I reject these submissions. I do not consider that the average consumer would perceive the photographer’s name as part of the Sign (as opposed to merely the context in which it appears) and none of the evidence from real life users suggests the contrary. In such circumstances the fact that the photographer’s name is garbled (while potentially relevant to context and addressed above) is irrelevant to the assessment of identity between Mark and Sign.

The clearest Getty Images watermark* produced in real life is on the First Japanese Temple Garden Image. Getty Images contends that, once again, (for both the figurative and the word Mark) there is clear identity in every respect: aurally, visually and conceptually. While it accepts that there is a difference because of “the presence of a shadow in between ‘a’ and ‘g’” in the word “images”, it nonetheless contends that this shadow is “hardly discernible and may form part of the underlying image”.

Looking at the watermark* as it appears on my screen (i.e. in its proper context on the image), I disagree. As Stability submits, the watermark* on the First Japanese Temple Garden Image appears to have an extra ‘i’ in the word ‘images’ so that it appears to read “imaiges”. The extra ‘i’ is no less “shadowy” than the other letters in the word and it is plainly not a part of the underlying image. It is not merely a difference in typeface or script. On balance I consider that this difference is not so insignificant that it may go unnoticed by an average consumer exercising a moderate degree of attention. The presence of the additional ‘i’ affects the aural analysis, just as it affects the visual analysis.

In closing, Stability suggested that I should conduct the comparison between Mark and Sign by reference only to Signs appearing in Annex 8I in relation to the relevant Model in issue and it produced in its skeleton the only other six Getty Images watermarks* included in Annex 8I:

I bear in mind Getty Images’ criticism that these were artificially ‘blown up’ and presented out of context, but, to my mind, what is clear is that none of these alternative Signs is identical to the Marks and I do not consider that they would be regarded as identical by an average consumer. The top left watermark* was affixed to the Alien Landscape Image to which I have already referred and is so blurred that even Getty Images conceded in closing that it cannot be relied upon. The remaining five all include obvious additional letters, or (in one case) an ‘m’ with three arches. On balance, none of these things is so insignificant that it would go unnoticed by the average consumer. A couple would come within the category of “nonsensical” watermarks* referred to by Mr Auerhahn. Although Getty Images stated in opening that, in the event that Stability sought to rely on any other examples it would address them in closing, I did not understand it to make any observations about any of these additional examples.

I agree with Stability that, in the circumstances, Getty Images has not relied upon a single example of a representative Sign (bearing a Getty Images watermark*) which is identical to the Marks in issue for any of the Models in respect of which I have found that a watermark* will have been generated in real life by a user in the UK.

In closing, Ms Bowhill submitted that if the court was going to look at watermarked images going outside the four images ordered by the court, it was only fair that it should do so having regard (also) to all of the Signs in Annex 8H (which runs to 101 pages). However, I do not consider this to be appropriate. I have held that there is no evidence of a real world user making use of verbatim, re-worded or “inspired” prompts and so it would be wrong to look at these.

There is no need to look at other Signs generated using the “vector art” prompt as I have found identity having regard to the examples chosen by Getty Images.

That leaves only images generated using the words “news photo”. Getty Images made no attempt at trial to identify any watermarks* on these images on which they specifically sought to rely for this purpose and I cannot see that it is for the court to proceed to carry out the necessary comparison without appropriate submissions from each side. This is particularly so where, as I have said above, the enquiry as to use is highly fact sensitive and depends upon an analysis of both the synthetic image and the watermark* appearing on that image. If Getty Images had wished to draw my attention specifically to any of the images in Annex 8H as an alternative to their chosen images, they could have done so in closing. Following circulation of the draft judgment, Getty Images suggested that the failure in the judgment to consider all other “news photo” images bearing the Getty Images watermark* in Annex 8H was a material omission to which they had a duty to draw my attention. They invited me to consider 17 additional images in Annex 8H in connection with their claims under sections 10(1) and 10(2) TMA. Alternatively they invited me to permit both parties to make further submissions on the point. Having considered the matter with care, I decline to take this course. Given that Getty Images had every opportunity at trial to make submissions about any of the images in Annex 8H, I cannot see that it is an obvious omission that this judgment does not deal with images in respect of which they chose to make no submissions whatsoever (with the obvious consequence that Stability also made no submissions on those images). I do not consider a very general submission made in closing to the effect that I should consider images in Annex 8H (as recorded in paragraph 419 above) to be sufficient. It is far too late to invite the court after circulation of the judgment in draft to consider evidence in respect of which no submissions were made at trial. In a trade mark case involving comparison between mark and sign (and particularly in this case where all signs are rendered differently and appear on synthetic images of many different types), the court will expect to receive proper submissions on any relevant sign upon which reliance is placed for the purposes of the comparison. It is unsatisfactory and inappropriate to expect the court to undertake such an exercise in the absence of any submissions whatever (including by way of identification of any specific images on which reliance is placed). That a party may appreciate too late that it should have focused its attention elsewhere in the evidence is not a reason to consider that an omission has been made in the judgment and nor is it a reason to permit a party to re-open issues that have been fully argued at trial. It has been said before that the trial is not a dress rehearsal and that all and any arguments that a party wishes to make should be made at the trial. Getty Images had every opportunity to make out their case at trial and it would not now be consistent with the overriding objective to permit them to re-open that case.

I have found identity of Mark and Sign in relation to the ISTOCK Marks. There is, however, no evidence of any real world use of a sign which is identical to the Getty Images Marks.

Getty Images pleads that Stability has used the Signs in relation to “synthetic image outputs” and “the provision of synthetic image outputs” which are identical goods and services to goods and services specified in classes identified in each of the individual Marks. I consider this topic for both the Getty Images Marks and the ISTOCK Marks, notwithstanding that, if I am right to dismiss the identity of Mark and Sign in relation to the Getty Images watermark* examples, then this analysis is strictly unnecessary in respect of the use of that sign in those examples for the purposes of the section 10(1) TMA claim.

Although Stability accepts similarity between its goods and services and those registered for the Marks, it denies that they are identical. This is primarily because it contends that none of the goods and services for which the Getty Images Marks are registered can be construed as at the date of registration in such a way as to be identical to Stability’s goods and services – this, it says would be an impermissible exercise in hindsight. It submits that although the language of the specification may be very general, its scope cannot be construed to cover additional activities unforeseen at the time it was filed.

I begin by considering some general points raised by the parties.

Stability accepts that the words “synthetic image outputs” are an appropriate characterisation of the intangible outputs of the Stable Diffusion Model. In closing, Getty Images appeared to suggest that Stability’s goods might in fact be characterised as “photographs” on the basis that that is how some users would refer to the outputs of the Model. I reject that case. It is not pleaded and it is far too late to advance a new characterisation of the goods in question only in closing. A synthetically generated image is not a photograph and I do not accept that it would be regarded as such for the purposes of trade. The Model is certainly not taking or composing photographs in the literal meaning of those words. That seems to me to be so even if the words ‘photo’ or ‘photograph’ might sometimes be used (wrongly) to describe such an image (see by analogy British Sugar – jam is not a dessert sauce even though “it too can be used on a dessert and everyone knows and sometimes does this” (at page 289, per Jacob J)).

Stability contends that the phrase “provision of synthetic image outputs” is “legalese” and does not reflect real world usage. It submits that such language fails to recognise that the user of the service is guiding the generation of images and controlling the contents of the outputs using their individual prompts. It says that a distinction must be drawn between the various access mechanisms: when the Model is downloaded from GitHub and Hugging Face, the user gets an AI image generator in the form of source code and model weights which are intangible goods, whereas, by contrast, DreamStudio and the Developer Platform are Generative AI services.

The question of whether the code and model weights are “intangible goods” is a question to which I shall need to return in the context of the case on secondary infringement of copyright. I do not understand it to be a question that I need to determine in connection with this issue, and, in any event, the code and model weights are different “goods” from the “synthetic image outputs” pleaded by Getty Images and accepted by Stability as a fair characterisation of the Stability outputs. Getty Images does not plead or pursue a case that the code and model weights made available to users for the purposes of download to a local computer is an identical (or similar) “service”.

Getty Images does not distinguish between the different access mechanisms, essentially because its case is that for each mechanism there is both the provision of the identical service and the provision of identical goods. I accept for DreamStudio and the Developer Platform (for reasons explained in more detail below) there is both. However, I agree with Stability that the real nature of the service in relation to the ‘download’ mechanism is releasing the code and model weights to users. I therefore agree with Stability that there is no identity between this service and the relevant specifications upon which Getty Images relies. However, to my mind, the goods then generated by the Model are synthetic image outputs – and this applies to all access mechanisms.

I reject Stability’s case that the word ‘provision’ is legalese and does not reflect real world usage. I have already made findings as to Stability’s level of control and these findings are also relevant here. By releasing the Models, Stability has made them (or more accurately the software which makes up the Model) available to the public to use – specifically, for this purpose, via the DreamStudio platform and via an API in the form of the Developer Platform. It is clear from the Expert evidence that the Model does not itself store images and so does not “provide” the user with an image that already exists – instead it generates a new image. But, although it is obviously more accurate to say that the Model generates synthetic images, I do not consider it to be unrealistic or unrepresentative of real world use to infer that many consumers will regard the generation of synthetic images by DreamStudio or the Developer Platform as involving “the provision” of such outputs. That seems to me to be particularly the case where, as Stability accepts, DreamStudio and the Developer Platform are themselves services. Furthermore, as Ms Bowhill points out (and I agree), the act of creation is not an act of trade mark infringement and so this is something of an artificial point in any event. The court must focus on the acts which constitute trade mark use – here the affixing of the Sign to an image.

Getty Images plead that these goods (synthetic image outputs) and services (the provision of synthetic image outputs) are identical to various of the goods and services for which the Marks are registered. Their best case on this was first identified in a letter dated 28 May 2025 and then set out in tables in their opening submissions dealing with the Getty Images Marks and the ISTOCK Marks as follows:

Without limiting her clients’ case, Ms Lane submitted in closing that synthetic image outputs and/or the provision of synthetic image outputs plainly fall within:

“(i)

digital media, digital materials, digital content, … namely images…and the above mentioned goods stored or recorded on electronic or computer media or downloadable from databases or other facilities provided over global computer networks” (class 9); (ii) “electronic delivery of images…via a global computer network and other computer networks” (class 38); (iii) “digital imaging services” (class 41); and (iv) “downloadable digital illustrations and graphics” (class 9).

She did not attempt to justify any of the other pleaded specifications.

I am bound to say that Stability’s oral submissions on this issue in closing were a great deal more extensive than its written submissions. Many points were raised for the first time without notice to Getty Images and often appeared to be straying into the territory of questioning the validity of the specifications – not an issue in these proceedings. If Stability intended to take wide-ranging points about the construction of the specification at trial, I consider that it should have alerted Getty Images to those points much earlier than in its oral closing submissions. Furthermore, in so far as Stability sought to distinguish in those submissions between various access mechanisms, it should also have made that clear much earlier.

Notwithstanding Mr Cronan’s engaging submissions, I have not always found it easy to follow how it is said that the specifications apply differently to different access mechanisms. My task was not made any easier by the fact that Getty Images (no doubt because they had not understood the extent of the dispute on this topic) dealt with the issue of double identity with what might be described as an extremely light touch, often not seeking to explain what individual specifications meant, but instead relying on little more than bald assertions that those specifications covered the goods and services in respect of which Stability had used the signs.

Owing to Getty Images’ continuing case that I must look at all of the goods and services relied upon against each Mark (and not simply the goods and services on which they focused in closing), I deal with them all below. In so doing, I also address Stability’s arguments, as I understood them, as best I can.

I do not consider the provision of synthetic image outputs to be identical to “photocomposing services” (class 40). The entire extract relied upon in the Mark from Class 40 reads “Treatment of materials; Bookbinding; framing of works of art; lithographic printing; photocomposing services; photographic printing; Printing”. Stability provided me in closing with an extract from the Encyclopaedia Britannica identifying the meaning of photocomposition as a “method of assembling or setting type by photographing characters on film from which printing plates are made. The characters are developed as photographic positives on film or light sensitive paper from a negative master containing all the characters; the film, carrying the completed text, is then used for making a plate for gravure or lithographic printing by a photomechanical process”. Bearing in mind this definition (which I did not understand to be disputed by Getty Images) and its context in the specification, it seems to me that Getty Images was right not to rely specifically upon this in its oral closing submissions.

However, on balance I consider the provision of synthetic image outputs (on DreamStudio and via the API) to fall within the words “digital imaging services” (class 41) and to be consistent with their core meaning. In closing Getty Images made no attempt to explain what this term might mean or why it was identical in the case of one or more of the access mechanisms. Stability also produced nothing designed to assist on the meaning of the words in the specification. Mr Cronan submitted that “digital imaging services” are vague words which, as at 2003 (i.e. the date of registration of the Mark) would have meant editing and processing of digital images – although he produced nothing in support of this submission.

Doing the best I can, it seems to me that there is a lack of precision in the term “digital imaging services” at least in the sense that it is not clear what such services might cover. Attempting to confine it to the core or substance of its possible meaning (see SkyKick at [365]) appears to me to involve construing it to mean the creation of a digital representation of the visual characteristics of an object, as well as the editing and processing services suggested by Stability. No doubt the development of Stable Diffusion provides a new means of achieving this, but the core service appears to me to be the same: the use of software online (DreamStudio) or via an API to facilitate the creation of digital images. Indeed this appears to me to be an example of “a new variant” of a service which falls within the specification, of the type envisaged in Reed v Reed. It is not a different service (cf. the decision in SkyKick to the effect that Cloud Migration is not an electronic mail service). I therefore consider there to be identity in relation to this phrase in so far as access via DreamStudio and the API is concerned. Downloading the Model via GitHub and Hugging Face appears to me to be somewhat different because users are able to generate the images locally and the service is in fact the provision of access to the Model in the first place, thereby enabling the user to generate images locally.

I consider that synthetic image outputs of the “vector art” type fall within the words “downloadable digital illustrations and graphics” (class 9). Even Stability accepts (in its written submissions) that these words are clear and that such “vector art” type 2D image outputs “are very close to being referrable as digital illustrations”, albeit that it suggests that “on balance, it would not be normal use of language in relation to these models to say “I generated a digital illustration””. In closing, Stability sought to run an additional argument to the effect that a full extract from the specification reads “…downloadable digital photographs, illustrations, audio clips, video clips, fonts, code snippets and graphics…”, submitting that none of digital “photographs”, “illustrations” or “graphics” is identical to a synthetic image output. On balance I disagree. I do not consider the specification to include ambiguous terms and I consider that the average consumer might well say that he or she had generated a digital illustration when describing an output image generated by Stable Diffusion of the type to which the iStock sign has been affixed. I therefore consider there to be identity between this specification and use of the Signs generated by each of the three access mechanisms.

Turning to the Getty Images Marks:

I take first “digital media, digital materials, digital content, … namely images…and the above mentioned goods stored or recorded on electronic or computer media or downloadable from databases or other facilities provided over global computer networks…” (class 9). Once again, Getty Images advance no clear description of what this is to be taken to mean, asserting in their closing submissions merely that synthetic output images fall “very easily” into this class. Stability submits that “digital media” and “images” are vague terms, which, pared back to their core meaning as at the date of registration, can only include stock photographs, editorial photographs and illustrations. I tend to disagree. I do not consider “digital media” or “images” to be vague or ambiguous and it seems to me that these terms plainly cover synthetic image outputs.

Stability also seeks to construe the words “other facilities” by reference to the word “database”, contending that “other facilities” must therefore mean “some kind of pre-arranged collection of content on-line” and that this must really be describing “how a stock content website works in 2012” when the Mark was registered. Again I disagree. While an AI model is not a “database”, it has been trained from a database and is another “facility” provided over a global computer network.

Finally, Stability seeks to differentiate between the different access mechanisms by reference to the concepts of “storage” and “downloading”. There is nothing in these points. In DreamStudio, files are stored on a server after generation (as Stability accepts) and they are also downloadable “from databases or other facilities provided over global computer networks”. When using the API, images are provided to the user via an input/output service (effectively, as I understand it, a means of making images available to the user without requiring a “download”) and when using the Model on a local computer (having downloaded it from GitHub and Hugging Face), images are generated on the local computer where they may be “stored or recorded”. Synthetic output images generated by each of the three access mechanisms thus appear to me to be new variants of the goods described in the specification (Reed v Reed).

I reject the proposition that the provision of synthetic image outputs falls within “computerized on-line search and retrieval services for images … photographs …still images…news images” (class 42). The provision of new images using text prompts does not appear to me to be an aspect or feature of a search or retrieval service for existing images (see SkyKick at [366]).

Getty Images say that the provision of synthetic image outputs also falls within “electronic delivery of images…via a global computer network or other computer networks” (class 38). Class 38 is concerned mainly with services that allow communication together with the broadcasting and “transmission of data”. The Explanatory Note to Class 38 states that the class does not include “content or subject matter that may be contained in the communication activity…”. I agree with Stability that this class therefore cannot be identical to synthetic image outputs, because it is not concerned with goods, but I consider that it is plainly capable of encompassing the provision of synthetic image outputs using the DreamStudio platform and the Developer Platform – both of which involve the electronic delivery of images via a computer network. To my mind, this specification cannot, however, encompass the ‘download’ access mechanism because, when the Model is being used on a local computer, the synthetic output images are being generated locally – they are not being delivered via a global network or any other computer network.

For reasons set out above, I consider that the provision of synthetic image outputs are identical to “digital imaging services” (class 41) when they are being generated by the user on the DreamStudio platform or via the Developer Platform.

As identified in the table above, Getty Images pleads a number of other specifications in relation to this Mark, under classes 9, 16, 35, 38, 41 and 42. Dealing with these as concisely as possible:

Getty Images made no attempt in closing to explain how any of the specifications referred to above might be identical and, accordingly, I reject its case on identity in respect of these specifications.

Again there appears to me to be identity between “digital imaging services” (class 41) and the provision of synthetic output images by the DreamStudio platform and via the Developer Platform for the reasons set out above.

I reject Getty Images’ case of identity in respect of “photographs” (class 16), again for reasons I have given.

For the reasons set out above there is:

As appears from the above analysis. Getty Images continued to rely at trial on a number of specifications for various of the Marks in respect of which they advanced no positive case. I consider these should properly have been abandoned. It is neither desirable nor consistent with the overriding objective to require the court to consider parts of a specification absent any properly articulated case.

For the reasons set out above: