top of page
Writer's pictureJon Sharples

Style counsel: can artists stop AI stealing their style?

Updated: Aug 1, 2023

Jon Sharples, Associate, Howard Kennedy LLP


Source: court documents
At the start of 2012, His Honour Judge Birss QC (as he then was) handed down his judgment in Temple Island Collections Ltd v New English Teas Ltd (also known as the "Red Bus Case") and found that the copyright in the claimant's image of a red bus crossing Westminster Bridge, created "using a well known standard piece of software called Photoshop", had been infringed by the defendant's attempt at recreating the same idea. That decision created something of a furore at the time, with many commentators expressing dismay that the judgment had introduced uncertainty as to the application of one of the fundamental tenets of copyright law, namely that copyright does not protect ideas themselves, only the expression of ideas.

What an altogether simpler and more innocent time that seems now. A few weeks ago, Adobe announced that the latest edition of its Photoshop software now comes with its new "Generative Fill" tool, which uses generative AI to enable its users to "add, extend, or remove content from [their] images non-destructively, using simple text prompts to achieve realistic results […] in seconds". An early eye-catching experiment with this tool was the use AI educator Kody Young made of it to generate expanded backgrounds for some of art history's most famous paintings.

Whilst works by Leonardo and Botticelli have been in the public domain for over 400 years, it is not difficult to see why the ease and speed with which generative AI tools can learn and passably replicate artists' distinctive styles is understandably giving living artists and those managing the estates of artists some cause for concern.

In January this year, three artists – Nashville-based illustrator Kelly McKernan, Oregon-based cartoonist Sarah Andersen, and San Francisco-based illustrator Karla Ortiz – filed a joint complaint in California against the creators of AI image generators Stability Diffusion and Midjourney, and the image-sharing platform DeviantArt, which recently created its own AI art generator, DreamUp.

The process by which AI “learns” to generate works that emulate human creativity relies on having access and analysing large numbers of existing works, learning patterns to create its own versions. At the heart of the three artists' complaint is the fact that the image generators in question were trained on the LAION dataset of 5.6 billion images scraped from the internet without permission by a German-based not-for-profit. The images of most digital artists (including the three plaintiffs) are included in the LAION dataset, given that DeviantArt and ArtStation were lifted in their entirety, along with Pinterest and Getty Images, who have filed their own lawsuits against London-based Stability AI in both the Delaware federal court and in the High Court in London.

Perhaps tellingly, in the 46-page complaint filed by McKernan, Andersen and Ortiz, there isn't a single example of a comparison between one of the artists' works, and an allegedly infringing work generated by the image generators. In other words, at this stage it is a case about inputs – the way the AI models use copies of works to analyse them – rather than outputs. The extent to which the acts of accessing, reading, analysing, and mining data without a licence are themselves acts of copyright infringement, or whether they should be allowed under fair dealing, fair use, or other exceptions or defences to copyright infringement, is currently a hot topic of debate amongst lawmakers and lobbyists in the UK, EU and the US, in addition to being the primary subject of the lawsuits already mentioned.

Parking the important question of inputs and how models are trained for now (and perhaps for a future blog post), let's to turn to the question of outputs and whether copyright or other intellectual property rights are likely to protect an artist's style at all, with or without the involvement of AI. Copyright proceeds by asking if one image reproduces a substantial part of one, single, earlier image – if it does, and there is no applicable licence, defence or exception, it infringes.
Looking back at the Red Bus Case, it does appear that the judge considered that some elements of what might be called the "style" of the original image formed part of the "substantial part" of that image that he found to have been reproduced in the second. In particular, the judge focused on the "visual contrast features" of the original image, one said to be the contrast between "the bright red bus and the monochrome background" and the other the contrast between "the blank white sky and the rest of the photograph". However, it is reasonably clear from the judgment that it was only the combination of these "style" or "visual contrast" characteristics and their specific application to the same subject matter and highly similar composition that led to the finding of infringement. Had that same "style" been applied to another scene, with a different composition and subject matter, the reproduction of the style would have not have been enough on its own to constitute the reproduction of a substantial part of the original image. In the battle of style and substance, copyright law appears reluctant to treat the former as if it were the latter.

Within a few years of the Red Bus Case, in 2014, UK copyright gained some new copyright exceptions, including fair dealing for the purpose of "pastiche". That concept isn't defined in law and has received very little judicial treatment since, but in its guidance published in 2014, the UK Intellectual Property Office offered two possible definitions, the first being a "musical or other composition made up of selections from various sources" – i.e. a collage – the second being a composition that "imitates the style of another artist or period". So now there is a whole new exception explicitly designed to permit imitations of artists' styles… provided they are "fair", leaving that subjective notion to do some heavy lifting, especially in a jurisdiction like the UK where is no tradition of evaluating fairness in relation to a tort of unfair competition. My presumption is that if "pastiche" had been in play in the Red Bus Case, Birss would not have found that the pastiche was fair in the particular circumstances of that dispute.

Given that "style" in an art context tends to refer to relatively subtle visual characteristics and/or patterns that are repeated across several works or even an entire oeuvre, it is clear that the 'one work vs one other work' nature of the test for copyright infringement means that copyright is always unlikely to protect style as such, with or without a fair dealing exception for imitation of style. This issue is currently playing out in an interesting case initiated last year in New York (sometimes it does feel like US art lawyers have all the fun…) by the internationally celebrated collage artist Deborah Roberts against the artist Lynthia Edwards and her gallerist, Richard Beavers, in which Roberts is accusing Edwards of creating works that are "substantially and confusingly similar" to her own works.
Source: court documents
On the copyright points, the motion to dismiss the case drafted by Edwards' lawyers presents a compelling art historical case to demonstrate that artists cannot reasonably expect a monopoly in "styles" generally and in mixed-media collage in particular: "Much like Pablo Picasso did not own the exclusive right to depict people in the cubist style (figs. 1-2), and Franz Klein [sic] did not own the exclusive right to paint abstract black marks inspired by ancient calligraphy (fig. 3), Deborah Roberts does not own and did not invent the subject of Black children nor the style of mixed-media collage." The history of art is mostly written as the history of concentrated periods of exchanging, borrowing and stealing ideas and styles under the banner of 'movements'.
Source: court documents
No doubt mindful of how challenging the copyright arguments are for their client, Roberts' lawyers have also advanced the case that Edwards' works infringe Roberts' "trade dress" in her works (something akin to "passing off" in English law), the features of which they list in her complaint as follows:
  • subject matter featuring Black children or adolescents, either individually or in small groupings

  • black-and-white or high-contrast photographic fragments of facial features, hair, arms, legs, hands and feet, assembled in angular, unnatural arrangements

  • expressive hand gestures

  • colorful, patterned fabric swatches

  • hand-drawn or hand-painted details

  • plain, unpatterned backgrounds

Seeing a lawyer's attempt to break down the features of a great artist's work in this curiously bloodless manner is a good illustration of why proprietorial arguments about stylistic features expressed at this level of generality are always unlikely to be convincing. Edwards' defence argues that even amongst those consumers who purportedly commented on the similarity between Edwards and Roberts' works, none was actually confused into buying an Edwards when they thought they were getting a Roberts, given all the different ways such confusion would be dispelled before the point of sale.

I have considerable sympathy for Roberts and for any artist who feels certain that some person or tool has deliberately set out to copy their distinctive style, and I understand from conversations with artists that it is a violation that hurts in a way that isn't entirely rational. There is more pressure than ever on artists to think of and position themselves as a distinctive 'brand'. However, I genuinely believe that what Roberts does cannot be replicated either by another artist or generative AI. The magic is in her specific visual taste, judgement of compositional balance, and in her lived experience and personal connection to her subject matter. In a recent opinion piece on AI in The Times, James Marriot offered the view that "human beings are principally fascinated not by brilliance but by other human beings" and went on to say that:

"Our fascination with other human beings is one of the most profound things about us. It transcends logic. As the psychologist Steven Pinker puts it, “we have deep intuitions about causal connections to people”, so that “a collector might pay $100,000 for John F Kennedy’s golf clubs even though they’re indistinguishable from any other golf clubs from that era”. To a non-human observer the pseudo-logic of that purchase represents a kind of magical thinking. To a human it makes sense."


And nowhere is the power of that "magical thinking" more keenly felt than in the art world.

Of course, if consumers are genuinely deceived into buying one thing when they are expecting another, the law should always step in and put a stop to that. But art is so rarely about the mechanical application of a set of rules or formula and, more often than not, copycats serve to show us what was so good about the original. It may yet be that artificial intelligence's greatest achievement is to remind us why actual intelligence is so sexy in the first place.

Comments


bottom of page