As AI-generated art continues to blur the boundaries between imitation and originality, questions around the copyrightability of artistic “style” have come into the spotlight. Shama Mahajan explores whether style can truly be separated from expression in the context of generative AI and examines how courts have approached this question over the years. Shama is an LL.M Candidate at National University of Singapore, pursuing her masters in Intellectual Property and Technology Law. [Long post ahead.]
‘Create an Image in ____ Style’: Is the Non-Copyrightability of Style a Dogmatic Convenience or Dichotomic Confusion?
By Shama Mahajan
“There are no style-less works of art, only works of art belonging to different, more or less complex stylistic traditions and conventions.”
~ Susan Sontag in On Style
When the Ghibli style images flooded the social-media platforms, the possibility of copyright questions was defended or rather deflected by the shibboleth of ‘style is not copyrightable’ argument. But is this justification as universal as its portrayed to be? Can style so definitively be extracted and separated from the copyrightable expression, that at no point will there be any copyright claim to it? This post argues that style cannot be separated from an expression to be considered as a non-copyrightable element per se especially in the Gen-AI context. Thus, to defend Gen-AI’s stylistic imitations, one must demand more than this shibboleth under the copyright law.
Images Speak Louder Than Words but Also Differently than Words
Even without the complexity of Gen-AI, courts across jurisdictions have displayed greater comfort in analysing copyrightability of texts rather than non-textual works like photographs, paintings, musical work etc. Let’s take the Rentmeester v. Nike case in the US where the 9th circuit admits that unlike novels, plays and motion pictures where, by way of filtration, elements can be distinguished as protectable expressions and unprotectable ideas, photographs can’t be similarly dissected.
Midst this dissection lies ‘style’ which under the existing copyright regimes in most jurisdictions is considered as unprotectable as it does not constitute an ‘identifiable’ expression. However, is this because it’s difficult to discern it within the expression as copyrightable or because it cannot be legally considered an expression, is where courts have struggled, swaying both ways.
Traditionally, copyright was envisaged from a writer’s perspective rather than a photographer, painter or an artist. As argued by Prof. Rebecca Tushnet, other non-literal works were added to the US copyright law more for economic reasons rather than conceptual demands and subsequently everything was attempted to be fit into the literary mode. By extending protection to other non-literary works, one can’t also extend the same yardstick of evaluation to all these works because, what constitutes their creative elements are different. The interaction with an image is visual hence more sensory driven rather than textual. Thus, a legal analysis of images under copyright law can’t be devoid of the artistic evaluation.
Though, Justice Holmes had cautioned the judiciary from becoming an arbiter of creativity when they were only trained in law but, scholars like Roland Barthes and Mikhail Bakhtin view art as a dialogue that derives meaning from artist and viewer interacting with the art. Thereby, in essence every judicial analysis is indeed influenced by the artistic value judgement even when they claim it to be an objective assessment. Appreciation of art theory can pave the way for a more nuanced legal analysis than merely an overarching claim of ‘style is non-copyrightable’.
Style: The Wordle of Interpretations
The more prevalent conceptual understanding of style tends to describe it as the manner of conveying the content of the art.
However, Susan Sontag in her essay ‘On Style’ highlights this very misplaced reliance on trying to separate style from content. Sontag argues that style is more intrinsic to the work rather than an outer layer, as it is embodied within art itself. Style is integral to an author’s personality and a way of resisting the homogeneity in art.
In essence, style is more than just a form, process, or system of classification of art. Within the artistic realm, style is not a closed definition but an augmenting concept that is agreed to be integral to the work of art.
Aesthetic Similarity: The Paradox of ‘seeing but don’t know what’
The two images above are different yet alike in what one would loosely term as ‘style’. My neighbor Totoro is a Ghibli anime, and what ChatGPT has produced is a ‘stylistic imitation’ of Studio Ghibli. Indeed, a dissection of the two will bring various dissimilarities to the forefront; however, one can also identify that certain elements are copied, as one can sense those similarities in the ‘look and feel’ of the images.
Let us now contrast the same with the anime comparisons above. All 3 are different anime studios that produce anime content. A viewer at first glance, when see the 3 pictures, identifies them as ‘anime’, which can be called ‘style’ in a broader term, yet can also differentiate Ghibli from Makoto in its pure visual sense and feel, again can be termed as ‘style’. To put it simply, a layman who lacks the vocabulary of an art critic can still identify Van Gogh’s ‘style’ in all his works, just like within anime, Ghibli can be distinguished even if the person can’t articulate what it is that allows him to draw those distinctions.
Thus, it would not be erroneous to say that, in authorial and personality arguments of copyright, ‘style’ is the personal expression and authorial element that is attached to the work by the artist/author. As pointed by Arjun Gupta, style is a sum of artistic decisions encompassing the conception of a subject and its expression by the artist. Thus, a prima facie Gestalt comparison of the works by a lay person is governed by the aesthetic similarity and places aesthetic intuition over analytical dissection.
Courts and Their Confusion with Style
Interestingly, despite the proclamation of non-copyrightability of style, courts acknowledge that the idea-expression distinction is not a water-tight framework to classify style in non-textual works. The two famous US cases in this regard are Fraklin Mint Corp. and Steinberg that treat style differently within the copyright framework, engaging in an analysis of aesthetic elements and ascribing it legal value.
In Franklin Mint, the subsequent commissioned painting of ‘the cardinals’ by the artist who had painted the original work was held to be non-infringing. In the analysis, the court dwells on ‘realism’ as a style which by itself grants limited copyright protection as it’s limited by the ‘reality’ of what is being depicted. Sobel and Tushnet both argue that this approach of understanding ‘style’ and its application is misguided. ‘Realism’ as a stylistic convention in art does not stem from the real truths of the world. But, the court’s obiter ‘the ease with which a copyright may be delineated may depend on the artist’s style’ is attempting to classify ‘styles’ which are ‘less associated/limited with real-world’ into the possibility of being protected as ‘expression’. In Steinberg, the New Yorker cover in which the artist depicted a myopic view of the world from a New Yorker’s point of view. The defendant’s movie poster for ‘Moscow on the Hudson’ represented Manhattan from the west with the lead cast of the movie. (see image below)
The court, in summary judgement held the defendants to be infringing the copyright of the plaintiff. The rationale was based on ‘stylistic imitation’ of the plaintiff’s artwork. The vantage point and the ‘sketchy, whimsical style’ were that of the plaintiff, which was copied.
Both judgements have received a fair share of criticisms; however, what is important to note is that courts often rely on ‘style’ while undertaking the substantial similarity analysis. What is not clear is in what way the ‘style’ will be treated. If the ‘non-copyrightability of style’ logic was to be applied, then in Steinberg, the fact that New York and Moscow are different expressions should have weighed more against the ‘stylistic similarity’. Bringing us to the most important question that legal analysis must ask is ‘to what extent is style not copyrightable?’.
Gen-AI thrives on what we can see but can’t define!
‘Style’ is more than just a process, for if it was, defining it as process/methods under copyright laws would not be so difficult. As far as Gen-AI is concerned, it is not detached from aesthetic sensibilities (irrespective of how it learns and generates/reinvents them) given the output is not just a factual extraction from the training data. Thereby, Gen-AI’s ability to replicate a style is not a result of ‘learning a process’. There is an intent to be able to use the ‘expressive style’ of an artist, which can be assumed if not ascertained from the LAION image-text dataset on which Stable Diffusion was trained. The text contained within these images was not limited to describing the object but also, the names of the artist and the school of art, which can’t be presumed to be serving no purpose at all, and it’s indeed not just limited to learning facts. For example, when it was announced that DALL-E 3 is designed to refuse generating output in the style of living artists, it seemed like an attempt to avoid plausible litigations and an admission that the latent space does embed the stylistic metadata of a creator, which is a personal expression that a creator confers upon his/her work. In a neighboring rights domain of ‘personality’, attributes of an individual’s persona are recognised and protected by the court, including in India, where a person’s voice and other attributes were held to contribute towards their personal identity and artistic expression. Thus, style in certain forms is considered integral to artistic expression. The question is to what extent is its imitation and reproduction are permitted under the copyright law.
Where Does India stand?
Style, even in India is not an expressly copyrightable expression. However, the recent injunction by the Delhi High Court in the case of Fashion Designer Rahul Mishra sheds some light on the fact that style is not entirely devoid of legal recognition. Though the High Court’s order does not clearly discuss whether style constitutes copyrightable expression, the subject matter of the lawsuit indicates that style is intertwined within its scope. The copyright infringement claims concerned certain motifs from Rahul Mishra’s Sundarbans collection that depicted the tigress, and it was alleged that the same constituted artistic work.
If we are to consider certain US precedents like Folkens v. Wyland and Enchant Christmas Light Maze & Market Ltd. v. Glowco, the copyright protection in depiction of nature, animals etc., is ‘thin’ which is to say that the threshold for finding of infringement is exact replication. By applying filtration, intrinsic, and extrinsic tests, it is likely that US courts might come to a different conclusion.
However, in this case, what the court has relied on is the ‘stylistic imitation’ by the defendant and copying the look and feel of the original copyrighted work. It is not about the tigress drinking water and creating a Sundarbans-like setting i.e. the idea, it is the stylistic execution of the idea in detailing the tigress, choosing to depict the nature in the same style that Rahul Mishra chose. Despite the absence of explicit acknowledgement, the court has, to an extent, recognised an artist’s stylistic choices as constituting part of the expression protectable under copyright. Thus, if AI creates an exact creation identical to that of the defendant, it will be held in breach of the copyright even though, for lack of better words, what is imitated is the ‘style’.
Conclusion
The aim here is not to state that style must be copyrighted, for it would create a legal absurdity and defeat the purpose of copyright law. The objective instead is to ask whether the law is able to appreciate the images and other non-textual copyright subject-matter for its nuances, or is it limiting the scope of these works by what can be reduced to text alone. Is the idea-expression dichotomy used as an alternative for that which cannot be legally understood or explained? Even before Gen-AI, the substantial similarity analysis for images did not clearly establish the position on evaluating ‘style’; however, it has not disregarded it. In the age of Gen-AI, where ‘what is taken’ is seen by law as ‘taking’, but the existing vocabulary falls short of describing what is being taken, it need not necessarily be a defence for that taking. Should lawyers now, as Prof. Tushnet suggested, consider familiarizing themselves with art to understand its nuances, not for the purpose of deciding its merits but for a better appreciation of ‘artistic expression’ which is as much a legal analysis as it’s of art? The answers may not be easy, but it also cannot be that ‘style is not copyrightable’.