Fashion AI: Cutting through the fabric of falsity

Spanish fast fashion giant Mango released its new ‘Mango Teen’ advertising campaign last month to much fanfare. At first glance, the advert – featuring a model wearing a colourful cropped top, posed against a desert backdrop – is similar to a plethora of other summer campaigns aimed at teenage girls. But upon closer scrutiny, something else becomes clear – none of it is real.

Mango’s campaign was created entirely with generative AI in a collaboration between its “design, art and styling, dataset and artificial intelligence (AI) model training, and photography studio, among others,” according to its press release. But while the industry celebrates its groundbreaking innovation, a host of lawsuits, regulations and consumer frustrations threaten to slow AI in its tracks.

That workers in the fashion industry are underpaid and mistreated is well documented, but the use of generative AI is compounding the problems. In 2023, Taiwanese-American model Shereen Wu discovered that a prominent fashion designer, Michael Costello, for whom she had modelled, had uploaded a digitally altered runway photo of her on his Instagram page. The issue? Wu had been altered to look white. To aggravate the situation, Wu had been reliant on the images from the fashion show to increase her exposure in the industry – she hadn’t even been paid by the designer. Now, she was completely unrecognisable.

Legally, models like Wu have few options in these situations. Copyright in an image is usually owned by the photographer and, as a result, it is only the photographer who is entitled to legal recourse under intellectual property (IP) laws if the work is altered without permission. The models, generally, do not have a say. It is also the photographers who have control of the images’ sale – they decide if the photographs can be sold and subsequently used in datasets to train and replicate in AI. This is also how fashion brands get away with using a model’s likeness to create an almost identical ‘AI twin’ without permission. These can then be used to create further campaigns and advertisements without paying additional compensation to the models, and often without their knowledge. Aside from the obvious problem of AI taking work away from models who traditionally rely on such campaigns for their income, the practice of replicating and reproducing a person’s image presents uncomfortable ethical issues.

It is ethical issues that are at the heart of growing consumer unease, particularly at the lack of clarity in how AI is used in design and marketing. Fashion brand Selkie, which markets itself as a ‘slow fashion’ company, found itself on the wrong side of a social media storm this year when it emerged that it was using AI to assist with designing collections. And since generating one AI image takes as much energy as fully charging a smartphone, it’s difficult to square with a message of sustainability. Levi’s suffered a similar backlash when last year it announced plans to use AI in order to generate models of more diverse body types and skin tones. Critics were understandably aggrieved that instead of paying a racially diverse group of models, the denim behemoth was instead creating fake versions. Others pointed out that the purpose of using models was to see how clothes fit on an actual human body – something that is impossible to achieve no matter how accurate the technology claims to be.

Fashion designers and artists are also struggling with the fallout from a shift to using generative AI in the industry. Lawsuits filed against fashion retail platform Shein highlight the frustrations of artists who claim that their work is being copied and reproduced by secret algorithms designed to scrape data. This, it is alleged, helps Shein to identify trends and reproduce copies of popular designs. While IP infringement claims are nothing new in the fashion industry, the use of AI to train algorithms and reproduce popular designs has added a complicated layer to an already problematic issue for designers.

Because of the speed at which generative AI has developed, legal systems have struggled to keep up with IP questions and ethical conundrums that have arisen. Finally, however, we are seeing some traction, at least in Europe. The European Artificial Intelligence Act, which came into force on August 1, will require brands operating within the EU to clearly label AI generated content so that customers are aware that AI has been leveraged. The Act will also require the disclosure of training data used to develop AI systems, which will, in theory, discourage brands from using images and designs protected by copyright law without the owner’s consent.

For models, however, the legal landscape remains murky, and only time will tell if AI will eradicate the profession or if consumer backlash and a desire to connect to something ‘real’ will triumph.