While I was immersed in the technology to write a book, I had a suspicion: this is the opposite of diversity.
The technology in question is virtual try-on. An online shop uploads a photograph of a garment, and an algorithm renders it onto models with different skin tones, body shapes and sizes. Retailers present this as a step forward for diversity: shoppers see clothes on bodies that look more like their own, whatever that body looks like.
Diversity, in the sense that matters here, means something narrower than an image. It means different people take part in something on equal terms. A model with dark skin stands in front of the camera on the same terms as everyone else and earns the same day rate. A model who wears a size 46 takes a job a smaller model would otherwise have got. Whoever looks different from the industry standard gets an equal place in the room.
Virtual try-on produces only the picture of that outcome. The casting and the pay that used to sit behind the picture drop out of the process entirely. A company can say it now shows diverse models on its website, and every one of those models can be an algorithm’s output. If the diversity on display is entirely synthetic, the company has redecorated its storefront while the casting and the pay behind it stay exactly as they were.
I was alone with this view at first. Most people who look at a rendered image ask only whether it looks right, and by that test it does. Retailers and shoppers agree on this, for a reason that has nothing to do with representation: conversion, the hardest currency in online retail, goes up. A shopper who sees a body like their own is more likely to buy, and the sale closes. Everyone involved has reason to call the result a success.
Fine. But what has actually changed on the other side of the screen? The same people keep making the same casting decisions, and every conversion figure that follows from a rendered image gives them one more reason to keep making them the same way.
Run the numbers and the AI version wins outright. A real model costs a day rate, travel, retouching and an image licence to negotiate. Rendering a body onto a garment costs roughly what it takes to write the prompt, and the question of image rights barely arises, because there was no photograph of a person to license in the first place. Judged as cost against benefit, synthetic diversity beats the real kind on both counts: wider apparent variety for a fraction of the spend.
What that calculation leaves out is what a booking has always been. Behind every casting sat a job, often several jobs across a season, and an income that gave a model a place in the economy the way any job does. A model who wears a size 46 and gets cast has rent covered by that booking. Render that same body instead, and there is no model left to receive the fee.
The same pattern shows up outside fashion. Companies elsewhere report the share of women or people of colour on their payroll as a claim about how they operate. That figure counts real jobs held by real people, which is why it shows up in a report at all. A rendered image never enters that count, and I doubt many companies have worked out that virtual try-on runs the identical manoeuvre through a different medium. The payroll behind it stays exactly where it was, even as the image on the screen changes.
Somewhere there is a model who could have had that job. Her face could have been the one on the website, and the fee would have gone to her.