Et in Arcadia ego: AI in modeling (ft. Lalaland.ai)
Beauty pageants are written into the public consciousness as a chaotic mixture of glamour, kindness and drama. Until, even that corporeal mixture, the sense of powder and sweat, and Chanel perfume disappear in the lines of binary code, as will be the case with the upcoming, first-of-a-kind pageant happening in Spring 2024: Miss AI. During a course of a lovely May evening, AI-generated models (sic!) will be judged by a panel of experts - two of whom are also AI-generated. The beauty queens are competing not only on beauty: their creators will also have to answer the beloved question of “How would you make the world a better place?”; then, they can win up to 13 thousand dollars.
Huxley couldn’t possibly make it up. Apart from the eerie organisation of the event, it signifies a change that is quickly approaching: AI models are beginning to appear at covers, in campaigns, lookbooks, or even shows. And this is not exactly news: in 2023, four different Vogues (edition Portugal, Italy, India and Poland) showed covers created with the help of AI - either bigger, or smaller; many other magazines (such as different editions of Glamour) were quickly to follow. In January 2024, Etro presented its first AI-generated campaign, too; however, Moncler and Valentino were faster, creating such lookbooks in the second half of 2023. Add to that ginormous sums of money that the AI-generated influencers earn (for example, Aitana Lopez is rumoured to gain 100,000 euros monthly), and the anthropogenic creatures created from machines, the Terminators haunting our dreams, from killer robots metamorphose into OnlyFans models. The questions are - who is creating those models? What does it mean for the modelling industry? And finally - in the world already obsessed with beauty and wealth, what will be the mental effects of standards becoming literally inhumane?
We are the Dreamers of Dreams: LaLaLand.ai and the illusion of algorithms
Before panicking, it needs to be addressed that digital influencers became popular, famous and rich even before the early 2023 ChatGPT-powered AI boom. Lil Miquela - the most famous example - came to „life” in 2016. Moreover, sometimes the distinction between a high-level CGI technology and AI is blurry. On top of that, many design programs, such as Photoshop, have included automatic fill functions based on AI years ago. What has changed, however, is the public awareness of these technologies and the amount of money and energy put into their development. According to McKinsey 2023’ The State of Fashion report, currently about 62% of fashion executives use GenAI technologies, especially GANs. This is an immense rise, as in the previous years those numbers were far below 40%. Moreover, if you would like to hire a digital model, no need to take a crash course in Machine Learning - you can just become a client of Lalaland.ai, a Dutch start-up, which build their expertise around rendering customisable digital models.
Laurens Houtekamer, Investment Coordinator at Lalaland.ai, explains that their technology can "provide fashion models to clothing brands and designers’. Instead of organising a model, a photographer, a location and a MUA, with LaLaLand’s software, you can take your pieces of clothing and within a few minutes, get pictures of your designs on any model. You can choose any model that you want, with customizable complexion, hair colour and facial features. „We are an addition to traditional photography”, further explains Houtekamer. “People use our product through the entire value chain and design processes - if they want to see how the design looks on a body. Also, they employ it for the end goal of e-commerce, and generate the on-model imagery, as we call it.”
Gender Shades and Datasets: Cascading Structures
It is not difficult to see that there are some potentially inflammatory points. Many experiments, most notably the Gender Shades project conducted by the MIT Media Lab, showed that AI algorithms are far from the goal of mathematical objectivity. In the now famous experiment, the researcher Joy Buolamwini was able to show that while facial recognition error for white men was 0.8%, for darker-skinned women it skyrocketed to 34.7%.
What is the reason behind those technical discrepancies? Multiple, but most notably from the dataset bias, stemming from human-added labels. However, as Houtekamer mentions, the company tries to tackle this by creating their own dataset - adding more diverse data. „In our field, which involves model imagery, there is a significantly larger volume of data for the conventional fashion look, characterised by thin, white models”, claims the Investment Coordinator. „Additionally, AI doesn’t necessarily generate new biases. Rather, it magnifies pre-existing human biases and makes them more noticeable. Previously, when tasks were completed by humans, these biases were not accurately documented due to the lack of precise records. Now, with AI, we can precisely track the choices made by the AI and mathematically quantify the extent of bias in an algorithm, thereby making the bias much more evident.”
Artificial Reality: A Sixth Sense?
Is it enough to ensure the algorithmic fairness and include HITL technologies? Neural-rendering is undoubtedly burdened with a tendency to homogenise and discriminate - and the laws are only adapting now. On March 13th 2024, the European Parliament implemented the „AI Act”, whose purpose is „to protect fundamental rights, democracy, the rule of law and environmental sustainability from high-risk AI”. What does it mean in practice? The biggest change is the creation of a ranking system which assigns different AI systems risks levels. For instance, certain Artificial Intelligence applications, such as certain biometric identification systems and preventive policing, were banned. Others, such as the use of AI in the critical national infrastructure, were tagged as „high-risk” and fell under tight regulations.
At the same time, the MEPs focused on the continuation of the establishment of Europe as a leader in this field. Logically, the innovations are going to accelerate regardless of laws - and the potential eugenics, while mortifying, are only one side of the problem. Ami Podebarac, a specialist AI technician at the London College of Fashion, mentions a more optimistic potential: that of AI to to create fashion in an accessible way. “This technology can be used to bring fashion to people who have a sight impairment”, says Podebarac. How does this look in practice?
Fashion technology advances have a long tradition of expanding the senses. One of the most innovative ideas is the SoundShirt: a wearable, shirt-like device created by a Swiss start-up CuteCircuit, which allows deaf people to feel music through vibrations of millions of small processors. The design of the SoundShirt started in 2002, and it was developed by 2016. And now - it could be even more. “The sensory technology can be massively improved with AI”, adds Podebarac. „For instance, it can give you the essential experience of feeling the texture that is actually digitalised.”
Models’ Inferno: 100 looks in 1 day
However, while the bias mitigation algorithms may aid in the dataset bias, and cutting-edge technology allows for sensory manipulation, social problems persist. The phenomenon of “digital blackface”, which was observed by media scholars years before the current AI boom, escapes the digital solutions. Defined as “the use by white people of digital depictions of Black or brown people or skin tones especially for the purpose of self-representation or self-expression” by the Miriam-Webster dictionary, it forces one to consider not only the final image synthesis, but also the beginning of the marketing chain. While one of the founders of Lalaland.ai is Black and mentions the need to morph the predominately white standards in fashion, the image synthesis technology created by the start-up is being used by many brands, not all of which have the most modern diversity standards. It is no longer the question of XAI, but of understanding the very much physical realities of the modelling market.
It is not only the BIPOC models who are facing loss of jobs: in general, models are becoming nervous. While it seems fairly obvious that designers muses - such as Shalom Harlow, Kate Moss, or more recently, Vittoria Ceretti or Bella Hadid - are indispensable, they are a drop in the ocean of talents who instead of working of editorials, need to pose for hours in luxury showrooms.
Maja Tyszecka, a Polish model working predominantly in China and Italy, has been modeling full-time for two years, appearing in Sandy Liang shows, Lamborghini commercials, and scoring multiple contracts with agencies all over the world. She considers herself successful - however, the realities of this job are harsh. „The part of my work that I find creative is maybe 20%”, says Tyszecka. „The rest is the uninspiring, an assembly line-like e-commerce nonsense. For 10 hours, you stand, changing outfits, and you have the same pose and the same expression. I've had moments where I did 100 outfits.”. AI can do it quicker and cheaper. As usually with changes, the superstars and the rich will become richer, and the already struggling will struggle more.
The simulacra in question: Abandon all hope?
Should we abandon all hope, collectively end the profession of a model, and let AI take Anna Wintour’s job? Well, the dominating emotions oscillate between excitement and dread. “After AI became such a mainstream point of everyone's life, all of us creatives became worried”, says Podebarac. “We are worried about the pure existential matter of what it means to create art.” The god of innovation has two faces: while AI can open doors to those who have previously been excluded due to their disability or class backgrounds, many skills, jobs and workplaces can be lost. As with web3, the economy of knowledge begins: the biggest currency seems to be creative, up to date and always en vogue.
Technoscepticism is crucial, but hope is needed as well. “I tend to be more positive in that aspect”, concludes Podebarac. ‘I think it's always been hard to be an artist.”
