H&M Knows Its AI Models Will Be Controversial

The first time model Vilma Sjöberg saw her digital twin, she found it both exciting and unsettling.

“It’s a picture of me, but it’s not me,” she said. “It was interesting how good it actually was.”

Even her boyfriend couldn’t tell the difference between a picture of the real her and the one created by AI, she said.

Exciting but unsettling has generally been the reaction to the use of AI models in marketing campaigns and on social media (that is, when consumers can detect the images aren’t real in the first place). H&M is betting digital dupes of real models such as Sjöberg will answer one of the thorniest questions facing fashion in the age of artificial intelligence: How do you take advantage of generative AI’s power to create realistic imagery while protecting the jobs and rights of the human models the industry has long depended on to sell its fantasies?

H&M’s solution is to bring the models into the process by producing digital twins which the models themselves own and control. It’s up to them to decide whether to let H&M use their twins — for a price, of course — in any AI-generated marketing. It wouldn’t just be H&M that gets access either. According to the company, because the models have full ownership of their virtual selves, they can make them available to any brand they choose, including H&M’s rivals.

The company plans to create digital twins of 30 models this year. It will likely introduce them first on social media and watermark the images to make clear they’re AI, in part to be transparent but also to see how people respond.

It knows not everyone will be happy.

“People will be divided. You know, ‘Is this good? Is this bad?‘” said Jörgen Andersson, chief creative officer of the Swedish fast-fashion retailer.

But in his view, AI is coming either way, and nobody is served by ignoring it.

“We saw that as a way, as a big player in the industry, to lead a conversation that takes the model, the agency and the best interest of the fashion industry into consideration,” Andersson said.

0 of 5

Since generative AI burst onto the fashion landscape, brands using the technology to save time and money, or just to experiment with a new creative tool, have faced anger and public backlash. AI’s detractors point to its developers’ misuse of copyrighted material to train their systems, its environmental impact and, importantly, the threat it poses to creative jobs, many of which already exist in a precarious state. Levi’s, for instance, ultimately shelved the plan it announced two years ago to use AI-generated models after facing a firestorm of criticism online.

Fashion and beauty businesses still believe that, with the right approach, they can use AI and assuage critics at the same time. The Estée Lauder Companies recently told The Business of Fashion that it is using AI for tasks such as editing imagery or generating pictures of obscure ingredients, but producing pictures of humans remains off limits.

H&M thinks by being up front about its use of AI and involving different stakeholders in the process, it can use AI models in a way that doesn’t alienate them or consumers. Andersson said the company worked directly with models and agents on its plan. Sjöberg, the first model H&M digitised, admitted she had been thinking a lot about the effect AI might have on her career.

But the company also acknowledges that it doesn’t have all the answers and needs to figure out details about how this approach will work. It doesn’t know all the impacts using digital twins will have on models, not to mention the other workers whose livelihoods could be upended — photographers, makeup artists, hair stylists, wardrobe stylists and others — if more fashion imagery is made by prompting an AI tool.

“It will affect the way we produce content potentially, but I can’t say exactly in what way,” said Louise Lundquist, a business developer at H&M.

How Do You Pay a Digital Twin?

Brands have for years now worked with so-called virtual influencers and models, but these digital avatars were typically personas unto themselves, not extensions of real people. They also still generally looked computer-generated.

Advances in AI are allowing companies and start-ups such as Botika and Lalaland to offer AI-generated models that are increasingly difficult to tell from the real thing. In an informal poll of BoF’s editorial staff, only a slight majority guessed correctly which of H&M’s images of model Mathilda Gvarliani was a photograph and which was made with AI. (In some of H&M’s other images the distinction is more obvious, though more because of the background than the virtual model.)

To create the twins, H&M takes numerous pictures of the models — in motion, from different angles and in different lighting — to capture them down to their birthmarks and movement patterns, making sure it has enough data to generate an accurate twin. It works with a technology partner called Uncut on the project.

The company sees the twins as a complement to their physical counterparts, not a replacement. Fashion companies churn out heaps of imagery for their marketing campaigns, e-commerce and social-media feeds. The content demands are key to why brands are eyeing generative AI. Now, a model could be walking a runway in New York while her twin is used to create an e-commerce lookbook in Stockholm.

“It’s a new revenue stream,” said Lundquist, who once worked as a model agent.

H&M is still exploring how compensation will work, but Lundquist said the deals will be structured like current arrangements. If a model does an e-commerce shoot, they’re paid for the usage rights to their image, which they or their agent negotiates based on the brand client, how many markets their image will appear in, the duration of the campaign and so on.

“This would be exactly the same,” Lundquist said. “It’s the digital twin being compensated for the usage rights of the digital twin.”

When companies have the chance to cut workers or reduce their pay, though, history shows they’ll take it. If part of the appeal of AI for brands is saving money, it’s hard to imagine there won’t be those that see an opportunity to slash the pay they offer.

“Would they pay more because they’re saving on the costs of travel, for example, or less because they don’t have to be physically present?” said Sara Ziff, executive director of the advocacy group Model Alliance.

Ziff pointed out that there are also legal questions about who owns the digital replica — the model or their agent — and which one the client negotiates and contracts with. While Lundquist said the models H&M works with have full ownership of their digital twins, other companies could be less scrupulous.

Historically, models have signed over power of attorney to their agents, allowing the agent to operate on the model’s behalf but also at times leaving models blind to their own business dealings, according to Ziff. It’s among the reasons that the Fashion Workers Act, a bill aimed at protecting models that takes effect in New York on June 19, has guidelines on digital replicas. It states that brand clients and agents have to get written consent to use a model’s digital twin and will have to detail the scope, purpose, rate of pay and duration of its use.

“Significantly, model management companies will have to obtain consent separate from the representation agreement,” Ziff said. “It can’t be buried in a larger representation agreement.”

A New Kind of Fashion Shoot

The rise of digital models could have other consequences. If a model can now be in two places at once, what happens to the model who would have booked the job the other had to pass on? Work could potentially become more concentrated among a smaller group of in-demand people. It could also change the nature of photoshoots themselves if you no longer need a full crew to create fashion imagery, with repercussions beyond the talent in front of the camera.

“I don’t love it,” said New York-based makeup artist Mary Irwin about the idea that more fashion imagery could soon be made by prompting a computer rather than via traditional shoots. She pointed out that most people on a set work behind the scenes, and it’s their combined work that creates the final result. “This collaborative, beautiful communal process that we have in fashion and in advertising, how is it possible if we’re not all [together]?” she said.

Hair and makeup artist Virna Smiraldi suspects she was recently involved in a shoot where the intention was to capture the models’ images for use in AI-generated imagery, but the client, a large e-commerce company, never specified the shoot’s purpose. Over the years, and especially since the pandemic, she said, companies have been trying to use fewer people on set to reduce costs. It’s become more common to ask models to do their own hair and makeup, for example. She believes the trend will only continue with AI.

“Maybe in very high-end or high fashion, there’s going to be a niche where the actual on-figure shooting is still going on, but for the rest of us, I think it’s going to be tough,” Smiraldi predicted.

Lundquist said H&M consulted with hair and makeup artists, photographers and other creatives, but she said she didn’t know all the impacts the adoption of AI by H&M and others would have.

There are many who don’t see AI as a problem in itself, including Ziff and Irwin. It all depends how it’s used. Andersson of H&M believes human creatives will remain indispensable. Without them directing the AI and pushing creative boundaries it’s useless, he said. It will only regurgitate what’s come before.

He also believes that, in time, people will not think twice about AI-generated imagery.

“AI is cutting across everything in the world today, and it will have effects on everything,” he said. “What effects we don’t know.”

His feeling is that the only way to find out is to start testing and see the results, for better and worse.