Fashion’s new fakes: How AI will change what you wear

Most experiments with Artificial Intelligence begin with a bit of faffing on a phone app. AI Arta or Wonder, for example. Tap in any crazy thing: “Old woman with tattoos nose-ring puffy rainbow hair and matching poodle...” and there she is. Instant gobsmackery delivered into the palm of your hand.

It feels miraculous that first time. Now watch an ad on the app (you don’t get nothin’ for nothin’ in 2023) and you can tap again: “Obese alien boy baby with powdered-wig waistcoat and angel wings” or whatever.

These are your AI “prompts”, instructing the app to target-harvest your concept from millions, possibly billions, of images already “scraped” from the internet. It scoops them according to caption (this is an important point) and mashes them into an idea of your idea. And there he is, alien angel boy. Amazing. Admittedly, a party trick that quickly wears thin but, still, amazing.

It’s a short leap to imagine this crazy, amazing AI thing slipping like a shiny new set of tools into the professional lives of visual creatives. Designers and photographers in the fashion industry, for example.

And so it has. Despite multiplying copyright court battles and increasingly dire predictions, of course it has. Who wouldn’t plunge up to their moral compass into a bucket full of such easy-peasily achievable miracles? Since AI apps started infesting phones and computers across the planet less than a year ago, the six creatives you’ll meet here have pushed their professional practices way beyond party tricks and into the seed phase of a new fashion industry that, by all their accounts, brings little shocks – pleasant and not-so-pleasant – with every new day.

Spanish designer Ines Aguilar, of slow fashion brand La Casita de Wendy, was an early AI adopter; she taught herself complex AI prompts and conjured 10 dreamy couture outfits that sold out (on pre-order) before their supernaturally lush patterns and piled-on floral embroideries were even cut and stitched into existence.

Melbourne-based photographer Philip Castle was another early AI adopter and, in one of those small, pleasantly shocking revelations common among creatives, says he “accidentally” designed a pod of vintage-style floral dresses – quite lovely and as viable for market as any if he can only source similar fabrics and get them made – while working on a folio of hybrid fashion images.

In the industry’s fresh “phygital” (physical/digital) space, Melbourne-born, London-based academic and hybrid designer Oscar Keene is also inventing his own AI tests, composing prompts to “ask” one algorithm (ChatGPT) how best to prompt another algorithm (MidJourney) to create his design concepts.

In fashion education it’s taken just a few thrilling months for AI to become a reliable teaching tool too, according to academic Todd Heggie. Heggie, part of the team at Melbourne’s LCI design institute, says it’s accelerated fashion course learning by several weeks and – bonus – freed precious time for students to practise and apply finer couture crafts such as French embroidery, crochet and other embellishments to their collections.

AI’s most wondrous results, however, are in fashion’s storytelling realm and these pepper social media platforms. Punch “AI fashion” into any search portal and thousands of generative and hybrid (a mix of the photographer’s work smashed into new visual concepts with AI prompts) images of unlikely models wearing unlikely designs in unlikely locations erupt on screen.

Among them are Milan-based Italian photographer Eugenio Marongiu’s often elderly models, heavily tattoo-ed or with ice-cream clouds of multi-coloured hair and weirdly wonderful fairytale wardrobes. “I’m still trying to understand [AI’s] potential,” says Marongiu, who also posts under the Instagram pseudonym Katsukokoiso.ai. He is one of AI’s busiest novices, with a brace of brand collaborations, media coverage and a group exhibition in China already under his belt.

Melbourne photographer Liz Sunshine plays with a similar aesthetic to Marongiu’s weirdly perfect, smooth-as-a-sucked-lozenge AI images, but she interprets it purposefully, as a radical rejigging of beauty ideals for older women. Her AI hybrid posts, featuring impeccably groomed, head-turning seniors, have also clocked up millions of views on social media.

These self-taught pioneers of AI fashion are still learning the which, why and how of prompts to tap into AI systems – algorithms with names such as MidJourney, Stable Diffusion and DALL-E.

“It’s been a trip,” says Castle. “I used to fantasise about this kind of thing, being able to blend this face with this face, add a dash of this and a dash of that and AI does it. It’s incredible.”

Castle has run increasingly meticulous prompting experiments with San Francisco-based MidJourney, one of the most popular AI systems, ostensibly developed for artists and recognisable for its painterly, Wes-Anderson-style aesthetic. In a second, even more complex process, Castle then uploads his low-resolution MidJourney images into programs such as Topaz Gigapixel AI. And with slides and dials and strings of even fussier prompts, he refines these until even the fuzz on an AI model’s cheeks is discernible in a final high-resolution picture.

He’s conjured a folio of ethereally beautiful new fashion portraits this way, without hiring a single model or switching on a single studio light. He has also, incidentally, rebooted a dormant chunk of his archive, a treatise on American culture he never got around to finishing in the 1990s, into a contemporary exhibition project by mashing original landscapes and AI-generated portraits into evocative new hybrid pictures.

“It’s kind of like surfing,” Castle says of his AI adventure so far. “Like riding a beast but there’s also a substantial random factor involved. It’s all in the prompts. You have to be able to describe what you want and, at a certain point you’ll want to upgrade [to a professional subscription in the AI program] and go into stealth mode so other people can’t see your prompts.”

Creatives habitually refer to their way with prompts as recipes or formulas, or “X” factors.

A single word placed in the wrong spot can change everything.

Eugenio Marongiu

“If you don’t give AI the right recipe it’s not going to bake a good cake,” says Heggie. “I’m a staunch believer that to generate new, original and interesting design outcomes you shouldn’t reference fashion in any way,” Heggie explains. “You should start by building a concept that’s external from fashion ... when you get closest to what you want, I tell students to start drawing over that, start dissecting the elements of the AI.”

He cites the complicated case study of LCI student and emerging designer Mathilda Singara’s fashion project as an example. Her idea, to use MidJourney and a blast of lateral thinking to develop garments inspired by the exoskeletons of extinct insects, at first went awry. She quickly realised that prompting with the characteristics of exoskeletons’ furry, crunchy bits didn’t live up to her offbeat fashion concept.

“She said, ‘OK, why don’t we put in [prompts for] an iris from an eye, and a lime, because they’re the shapes and colours we want’. So we did. We put in images of eyes and sliced limes and the AI created all these amazing print concepts; we picked one, sent it off, scanned in the [garment] patterns, digitally aligned them and [Mathilda] cut and made them.” Eyes, limes and insect exoskeletons fused perfectly into a remarkable AI hybrid fashion concept.

It’s predicted that similarly original strings of targeted, meaningful and complex prompts will eventually be recognised legally in the realm of intellectual property. Already in the US, some companies are paying a fresh-shucked breed of non-technical professionals six-figure salaries to be their prompt engineers. Prompts are that precious.

“I’m of the opinion that prompts, especially in cases of professional/artistic use, should remain secret,” says Eugenio Marongiu. “It’s like a chef revealing his secret recipe or a photographer giving away the raw files of his shoots.”

He prompts his own AI images into existence with the same intuitive, artistic beat and process as his real-world work. “I move exactly as for a photographic shoot, but everything is at the written and mental level: the type of models, the location, lighting, outfits, makeup and hairstyle ... a single word placed in the wrong spot can change everything.”

Creatives liken it to relearning your own language. “Initially my prompts were simple; 10 words,” says Liz Sunshine. “But I became more precise about what I wanted to create; I found the words to describe a scene to a deeper degree, sometimes 100 or so words ... imagining women I would photograph in real life. I enjoyed seeing each subject’s story unfold ... Simple details, like skirts wrinkled from sitting or a woman with lumps and bumps, make the images feel more real.”

Sunshine’s experiments have rocketed on social media, with one image clocking more than 16 million views. “I’ve been using AI in all its forms for many years [but] in photography it feels new, creating images as a collaboration between imagination, bots and words instead of cameras and light.”

She’s sped up her art processes and – another one of those happy AI accidents – plunged into a book project that virtually opens a new cultural gateway and wouldn’t have been viable even a year ago. “I’m self-publishing a book that questions style as we age and presents a series of AI images that celebrate the visual qualities of ageing,” she says. “I’ve always found older people beautiful; their grey hair and deep lines tell the story of life ... yet both myself and the women around me feel differently, we feel an urgency to cover greys and anti-age at all costs. I hope this book offers a different narrative to the one that’s often [celebrated] online.”

In a month Sunshine created an astonishing 6500 base AI images for the book and edited them down to an exceptional 200. “That’s a lot less time that it may have taken me to find 200 subjects that met my brief and capture them in the real world,” she says.

Frustratingly, there’s one category of fashion miracles yet to be mastered by AI: the laborious work of converting an on-screen concept into patterns and production instructions. Todd Heggie hopes that as machine learning enables AI algorithms to accelerate their own evolution, it might be possible in a few years, if not months. “Tech packs and spec sheets,” he says, laughing. “I think AI has the same disdain for that as we do.”

Even Ines Aguilar’s marvellous “Artisanal Intelligence” couture collection is rare for its real-life outcomes. “We designed the collection with a specific colour palette in mind, along with the materials, embroideries, everything,” she says. Real-life fabrics, yarns, even crafted details, in other words, were preloaded into her prompts.

Accessible AI is such a young revolution, fast evolving, fraught with unknowables and predictions both glowing and apocalyptic. Arguments are multiplying around the morality of AI algorithms that have “scraped” hundreds of millions of existing images from the internet, all without permission, many – possibly even most – protected by copyright.

Tap in a prompt that includes “in the style of” a particular artist or photographer, as some fashion creatives do, and the resulting AI image’s originality is automatically arguable. Definitions of “fair use” are increasingly being tested in courts by creatives worried their oeuvre is being hollowed out. As AI evolves, so does unease, despite assurances that algorithms skim and mash, they don’t replicate unless specifically prompted to.

“I do feel conflicted by it,” says Oscar Keene. “When I started I used prompts like ‘inspired by’ and then thought; is this not plagiarism?”

The integrity and originality of his own work is a constant focus. “I love designing and when I share my AI work with other people their response is, ‘yes, it’s pretty, but it doesn’t have that thing, that point of view that you have’. I think my point of view is diluted when it becomes part of AI.”

Keene continues to experiment but is ever vigilant and intellectually wary as AI evolves. “Sometimes I think when I’m using it, ‘Am I facilitating my own obsolescence?’ I think it’s a great equaliser; people that aren’t trained in design are able to execute these creative visions ... it’s a good starting point, gets the ball rolling on the design process. I’ve thought it would have been good when I was [studying design at RMIT] in first or second year but now ... it’s a play tool. The more I’m interacting with it, the less meaningful it is to what I want to do, to make designs that, when you wear them, it means something to you.”

Follow the creatives:
Philip Castle: philipcastle.com; instagram.com/philip.castle
Oscar Keene: oscarkeene.com; instagram.com/oscar.keene.designs
Todd Heggie: instagram.com/t.a_creative_consulting; lcimelbourne.edu.au
Eugenio Marongiu: eugeniomarongiu.it; Instagram.com/eugeniomarongiu; Instagram.com/katsukokoiso.ai
Ines Aguilar: lacasitadewendy.com/en/; instagram.com/lacasitadewendy
Liz Sunshine: instagram.com/lizsunshine