The first AI Fashion Week kicked off yesterday, digitally on a dedicated app and physically at New York's Spring Studios: showcasing collections designed by AI fans from all over the world using AI text-to-image applications, the event offers a glimpse of the future of fashion.
The collections can be viewed by registering on the dedicated app that also allows you to vote for your favourite designer (note: you must leave your mobile phone number and vote from a mobile device).
Organisers hope to showcase through the designs on the platform the potential of AI technology, explore the endless possibilities of AI and give three winners (finalists will be announced in May) the chance to enter the Maison Meta x Revolve AI Fashion Incubator and get their collections produced.
So what's the verdict on this first AI Fashion Week? Well, there were a few bumps in this maiden voyage. Participants were asked to create collections featuring 15 to 30 looks on models walking a runway, but were given carte blanche. When the site and project launched, the fashion models created by AI applications still looked rather bizarre: some sprouted extra fingers or had elongated bodies with small heads and extremely angular cheekbones, features that made them look like extraterrestrial creatures from a stylish yet imaginary planet located in a galaxy far away.
Just a few weeks after the first participants passed step 1 of the process, Midjourney released a new version (v5), that provoked much excitements among AI text-to-image fans, but also caused a major setback to the event as organisers decided to extend deadlines and allow people to redesign their collections or come up with new ones.
The last few days were crucial with the AIFW team working hard to upload the 133 selected collections, even though only half of them eventually made it to Spring in New York.
One interesting thing to note is that the event proved that fashion genuinely pervades our lives: in total organisers received over 350 submissions, and this was just the first time the event took place.
Besides, the people who submitted their collections didn't necessarily have a fashion background: among them there are fashion students and designers, but there are also people with completely different careers and backgrounds who just love fashion or people who don't know anything about fashion but work in technology.
In the weeks that led to the event, AIFW23 Discord channels were abuzz with participants sharing their thoughts and collections and obsessively wondering if they would have made the final selection.
In a way, those ones with a fashion background were maybe more relaxed about the process than designers with a tech background who suddenly showed a great interest in taking part in this experimental fashion week.
Another interesting point is that the participants' ages varied greatly: quite a few appear to be in their twenties, but the thirties to forties brackets are also widely represented, something that makes you wonder if some of the more mature participants may be reconsidering careers, or may be dreaming of a more creative profession merging fashion and technology.
Digesting all the collections takes time: some collections look more polished and coherent and were created by people who work in the industry; others look less cohesive, but there are many imaginary locations and sets, from mushroom fantasylands to deserts, while futuristic backgrounds conjure up spaceships and natural worlds can be interpreted as blissful tributes to the Pachamama.
Some collections are extravagant, the result of an active human imagination combined with a powerful system; others lack the extravagance, but would be easily translatable into real clothes; and then there are designs that show potential, but will need a lot of experiments before being turned into real pieces.
Some designs and models were maybe created with previous version of Midjourney; in other cases designers must have added a tiny dose of "--c", the "chaos" factor that, with the proper variable, can bring in some distortions and unpredictable elements.
In many cases accessories seemed more desirable than clothes: the offer includes sleek synthetic materials forming parametric structures around the models' feet or extravagant sunglasses creating biomimicry-inspired visors that may become trendy on Mars in 50 years' time.
One thing must be noted: many participants opted for co-ed collections, there are just a few ones featuring only male models; womenswear seems to prevail, even though in many cases the designs are genderless.
Most collections feature extra images showing not just the accessories, but also the backstage, street styles and front rows. The "f-rows" can be particularly difficult for Artificial Intelligence: the system finds indeed still difficult to come up with credible images of crowds with multiple people. Yet, in most of these cases, the designers did a great job, with some front rows immersed in a pleasant cinematic atmosphere or even featuring clones of Anna Wintour.
There are some aspects, though, that must be taken into consideration: white and thin models prevail, which may mean that some designers didn't take diversity into consideration or that they just wrote the word "model" in their prompt and the AI systems they used provided them with a white model.
That said some designers rebalanced their runways offering a diverse cast of models, representative of a wide range of body types and skin tones. One designer also included two models on wheelchairs and a runway celebrated elderly models, with a distorted clone of Iris Apfel.
As highlighted in a previous post, diversity is a key issue on both real and digital runways and even AI text generator ChatGPT acknowledges the issue.
Required to provide an opinion about the risks of AI designed fashion collections, ChatGPT wrote: "One of the potential risks of using AI to design fashion collections is the risk of bias and lack of diversity. AI is only as good as the data it is trained on, and if the data is biased or lacks diversity, then the AI-designed fashion collection will also be biased and lack diversity."
"If the data used to train AI only represents a narrow range of body types, skin tones, or cultural backgrounds, the resulting fashion collection may not be inclusive or representative of a diverse population," ChatGPT continued.
"This can have negative consequences on society as fashion plays a significant role in shaping cultural perceptions and norms. Moreover, AI can also perpetuate existing biases in fashion, such as gendered fashion or stereotyped fashion designs. For instance, AI may be trained on data that associates certain colors, patterns, or styles with specific genders or cultural backgrounds, resulting in fashion collections that reinforce these biases."
The other problem is the derivation of some of the collections, it is indeed undeniable that some designs are reminiscent of real collections from other designers/fashion houses.
There's a whiff of Prada or Bottega Veneta here, a hint at Fendi there, and then assorted traces of Alexander McQueen, Marni, Etro, Dolce & Gabbana, Guo Pei, Elie Saab and Valentino.
Besides, a model in one of the collections wears a sort of headpiece creating a frozen splashed effect around her head, reminiscent of Stephen Jones' 1993 "Wash 'n' Go" hat, and of Iris van Herpen's S/S 2011 collection (View this photo). Van Herpen actually remains a strong reference in those collections with a strong futuristic yet ethereal edge.
It may be the case that some designers put into the system a pre-existing image and asked the AI system to work on that, while others may have used an original prompt containing words that pushed the system to harvest specific ideas from existing collections.
Required to provide an opinion about AI harvesting ideas and images from previous collections, ChatGPT states: "Copying in a fashion collection designed by AI is a potential risk, just as it is in traditional fashion design. AI uses data and algorithms to generate designs, which means that there is a possibility that the AI may unintentionally or intentionally copy existing designs or ideas. Moreover, AI can also learn from existing fashion designs and trends, which means that there is a risk of producing fashion collections that are too similar to existing designs."
Yet, AI systems do not necessarily copy and plagiarise an existing fashion design when creating a collection. As ChatGPT explains, "AI can be trained to generate designs that are original and unique. The level of copying in an AI-designed fashion collection depends on the data and algorithms used to train the AI. If the data used to train the AI includes a wide range of fashion designs and trends, the AI-generated designs are more likely to be original and unique."
That said, apart from connections and links, there are some extravagant moments of originality during this first AIFW, with footwear often looking more desirable than what you see on the market at the moment.
So, could AI text-to-image harm the fashion system? Well, we must try and be very honest with ourselves here: for the last 30 years, many houses and brands have been constantly reinventing and remixing previous designs. It is not a secret that up until 20-25 years ago, designers often visited obscure vintage shops in search of that special golden egg, a forgotten garment that, slightly altered, reinvented and replicated, had the potential to make a lot of money (this was the main modus operandi at Prada, then it expanded to other houses as well).
AI does exactly the same, only it remixes and recreates or recombines things with a higher level of efficiency and without the need of physically going to an obscure vintage shop.
In a nutshell, we may just argue that AI systems are optimising what humans have been doing in fashion in the last few decades.
Maybe, to detect better possible derivations from previous collections, the AIFW organisers will have to pick among the jury members judging the collections, also a fashion historian or a journalist with decades of experience, since such figures would be able to immediately spot possible references. Hopefully, next time they will do so.
Another aspect that demands attention is the issue of copyright as the organizers intend to transform the winning collections into physical collections. However, it is unclear who will possess the copyright for such collections, considering that the generated content through Artificial Intelligence is assumed to be in the public domain.
For the time being what's for sure is that this first AIFW was a learning process: if the participants (and even the organizers) were able to gain something from this, then it can already be considered a success. Who knows, maybe some of the participants working in fashion will decide to start focusing more on technology and vice versa, or maybe some of these AI designers will decide to start collaborating together and produce something even more extraordinary for the next event.
As for fashion designers and houses, underestimating or condemning Artificial Intelligence is definitely not the solution: AI is a tool that may lead to new possibilities, after all some designers and houses are already using it in their collections, advertising campaigns and during dedicated events. But maybe we should follow ChatGPT's advice: "The best approach may be a combination of human creativity and AI optimization, which can produce fashion collections that are both efficient and emotionally appealing."
Comments