Original photographs benefit from copyright protection. But what about a photographer's distinct style?
Like many teenagers, I was interested in fashion. But what truly captured my imagination was fashion photography - particularly the sort featured in magazines such as Vogue, i-D and L'Officiel. Even when I couldn't afford to get my hands on physical copies, I still spent hours trawling through incredible editorials and photoshoots because the shoots were uploaded to Livejournal, Tumblr, and other popular content-sharing websites of the early 2000's. (Instagram was still a few years away...)
I'm dating myself, here, but all this to say I know my photographers, and one of my all-time favourites is Peter Lindbergh (1944 - 2019). I was deep in my French New Wave fangirl era when I first discovered Lindbergh's work, and loved the way that he – like Truffaut and Godard – captured natural and evocative feminine beauty in black and white. Moreover, his portraits of 1990s supermodels such as Kate Moss, Helena Christensen, Christy Turlington and others captured the women's unique personalities, helping to elevate them to celebrity status in their own right.
Lindbergh's ability to make subjects appear both vulnerable and powerful make his style instantly recognisable. And without question, his work remains inspirational and influential today. But what happens when this inspiration becomes emulation, thanks to generative AI?
How does genAI replicate the style of famous artists?
Algorithms underpinning generative AI models like those in Midjourney and DALL-E study and learn from huge collections of audiovisual content and text known as "datasets". OpenAI, the company behind DALL-E and ChatGPT, claimed in 2022 that its DALL-E model was trained on a dataset of more than 250 million images.
Some of datasets are labelled, which means tags or descriptions on the file help the AI understand and learn from the images more effectively. But not all datasets are labelled: Midjourney and other advanced models can also identify patterns and features of content on their own, without relying on labels. This makes AI learning more broad and flexible, with sources coming from a wide variety of content. The more diverse and extensive the training data, the better the generated outputs.
Of course, what the AI ultimately creates relies heavily not only on the quality of the training data and the sophistication of the algorithm, but on the user's interaction with the service.
From the user's perspective, creating stunning digital art is relatively straightforward - you simply write a plain-text instruction ("prompt") into a text field, describing what they want the image to look like. To obtain good results, prompts often specify desired artistic mediums such as "mixed-media", "oil painting", "hyper-realistic photography", and so on. Some go so far as to stipulate technical requirements such as "Canon EOS 5D, 35mm lens, ISO 100" or "charcoal sketch on rough watercolor paper."
It's also not uncommon for prompts to mention specific artists for inspiration: Gustav Klimt and Egon Schiele are particularly popular choices, as shown above. Now, several works by Schiele and Klimt are in the public domain, so we can put copyright considerations to one side for those examples.
But things become very complicated when when content that is still under copyright protection is used in training data. That Lindbergh's photos were almost certainly used within datasets could constitute copyright infringement if those photos were not appropriately licensed, and much has been written on this ever-expanding issue and the many related lawsuits. Within the broader question of how genAI technology and copyright infringement, a particular point I've been thinking about lately is the resulting generative content, and the legal grey area of style and essence.
The inspiration to write this post came from this image below, which I stumbled upon while browsing Midjourney's feed. It's strikingly similar to Lindbergh's genuine style - but look at the prompt: it doesn't mention Lindbergh at all.
While it comes as no surprise that prompts mentioning "Peter Lindbergh" by name can produce images nearly indistinguishable from his work, even prompts that don't mention him yield incredibly similar results. This happens because the AI has learned to recognise and replicate the distinctive elements of Lindbergh’s style from its training data, which will include hundreds if not thousands of examples of his work. The AI’s algorithms will identify patterns and stylistic features that define Lindbergh's photographs, such as his use of lighting, composition, and subject matter, and apply these elements. In this case, this could mean prompts requesting "portrait" photos of "beautiful women" on "beaches" in "black and white".
AI, Aesthetics, and Copyright
Copyright attaches only to the expression of ideas that are fixed in a tangible medium, such as on paper or on a computer file. In other words, the actual creative expression contained in a photograph - the lighting, angle, composition, and other artistic choices made by the photographer - are copyright protected. What copyright law does not protect, however, is the style, technique, ideas or vibe behind the image.
That said, when an artist’s style is so distinctive that it becomes a recognisable signature - as with Lindbergh's work, I would argue - the boundaries become blurred.
What does this mean in practice? To determine copyright infringement under UK law, we must consider whether a "substantial part" of the original work has been copied. The test is qualitative and not quantitative, so infringement may occur even if just a small portion of the work was copied, provided that small portion is a significant part of the original creator's skill, labour, and expertise. (People who assume that you can avoid infringement by simply changing a particular number of elements in a given work are wrong, I'm afraid.)
But if the second photo merely emulates the general style or aesthetic of the original without directly copying the protectable elements of a specific work, it's unlikely to be considered infringement. So while the expression in individual photographs will receive copyright protection, a photographer's unique style itself cannot be directly copyrighted. In other words, adopting a similar aesthetic style is generally okay under copyright law.
There are several reasons why legislation takes this approach. Firstly, in the Anglo-American legal tradition, a fundamental principle is to promote an environment where artists can share, adapt, and be inspired by each other's work. Art, literature, design and music - all forms of human expression are iterative to some extent. This is also why copyright protection doesn't last forever, and why certain exceptions such as fair dealing and parody apply. Basically, law makers and the courts recognise that the free and dynamic exchange of ideas is essential for cultural and artistic development in society. If copyright protections are too strong, it would have a chilling effect on artistic expression. Secondly, in practical terms, defining an artist’s "style" would be highly subjective and challenging: styles are inherently fluid and evolving, making clear legal boundaries difficult to draw.
Herein lies the key question: is existing copyright law fit for purpose in the era of GenAI? Should a machine that can imitate human creativity be held to the same standards as a living, breathing photographer? Clearly, the law should not stifle the evolutionary nature of art by granting broad copyright protection to something as potentially vague as "style". On the other hand, failing to adequately protect nuanced and distinctive styles might undermine the very concept of originality and the intrinsic value of creativity. From a financial perspective, if AI produces work that is indistinguishable from an artist's genuine repertoire or portfolio, it could flood the market with similar works. This in turn might devale the original artist's creations and reduce demand for their "real" work, leaving creatives to compete with algorithms that mimic their art at a fraction of the cost.
Personally, I find the space where AI, aesthetics, and copyright meet to be fascinating. But fun and philosophical discussions about legal theory aside, these issues are complicated (and very much not "fun", particularly if you're in the middle of a lawsuit). Even as an "expert", I find it difficult to draw bright lines between inspiration and infringement. What I do know, however, is that while generative AI can produce stunning and evocative art, I think it's going to fundamentally change our understanding of creative ownership and the protection of artistic styles.
Genuine or GenAI? - Answers:
| The portraits of Kate Moss (left) and Alicia Vikander (right) are real Lindbergh photos. The one in the centre is a Midjourney creation. |
| The top left, bottom centre, and right (Kate Winslet) are genuine by Lindbergh. Bottom left and top centre are Midjourney versions. |
Comments