The Edit is our weekly show, where Social Editor Grace Robinson quizzes Editor-in-Chief Ben Hanson on five of the most significant fashion and technology stories from the past seven days. This edition covers the racial-discrimination complaint over AI-altered model imagery and Meta’s new Muse image generator; the community backlash against AI data centres in Scotland; Whering’s $7m raise to bring AI to the digital wardrobe; AI-powered sorting of textiles and secondhand clothing from Fleek and reverse.fashion; and OpenAI’s full-duplex GPT-Live voice mode.
This show is available in video format, embedded below, as well as audio, on Apple Podcasts and Spotify.
For deeper interviews with brand and technology leaders, our prestige interview show releases on Thursdays everywhere you find podcasts – and will be coming to YouTube soon.
Subscribers to our mailing list will receive notifications for both shows.
Grace Robinson: Welcome to The Edit from The Interline — the show where we run a quickfire analysis on our pick of the most important fashion and beauty technology stories from the last seven days. I’m Grace, the Social Editor, and I’m joined by Ben, the Editor-in-Chief. Together, we have less than twenty-five minutes to give you our analysis on the stories that we think really matter.
Ben Hanson: Hey, Grace. Good to see you again. It’s hot again — this is the second heatwave recording I think we’ve done, so we’re under time pressure and environmental pressure this time.
Grace Robinson: Yep, still powering through, but we’ve got a lot of stories today, so let’s get into it.
Ben Hanson: Go ahead, fire away, hit me.
AI likeness and consent — the Peter Jackson complaint and Meta Muse
Grace Robinson: So, as always, AI dominated the headlines this week, and the stories were quite polarising — this is going to be a very AI-heavy episode.
The first story I want to talk about actually relates to a couple of others we’ve covered on The Edit before. Just to remind everyone: one was the story about Francesca Pujos, the model who sued a retailer for using her likeness to generate new AI images without her consent. The other was about the outdoor retailer REI, which fell foul of Meta’s new AI ad programme because it didn’t opt out.
This week it’s another very similar story. A Nigerian-Australian model, Elii Emeghebo, has sued the retailer Peter Jackson, alleging that it used AI to alter his face — changing his skin tone and his eye structure — and he’s accusing the brand of racial discrimination through AI.
Equally concerning, it was also announced this week that Meta has revealed its new Muse AI image-generation model, which will let users generate images using the likeness of other Instagram users who don’t actively opt out.
To me, these stories highlight how many questions and unresolved issues there still are around how people use AI with commercial imagery, and how much of a grey area it remains. So I just wanted to get your take on what’s going on with all of this.
Ben Hanson: Yeah. I’ll do them together, in reverse order, I think. We are firmly in the ask-forgiveness-not-permission stage of the AI rollout. You mentioned the REI example from last week, and the Francesca Pujos one. Those are both instances of companies of very different scales — from a fairly small-to-mid-market retailer up to Meta, one of the world’s biggest tech companies — effectively saying AI is carte blanche, a licence to do what we want. The frontier is untested. The legal precedents are not set. So let’s take our chances and play around with these things.
Now, Meta in particular seem committed to running down the evil-genius playbook at the minute. If your brand is as out of favour as Meta’s is, part of me understands treating that as an opportunity — to say, let’s try to get away with as many things as we can, because people don’t like us anyway. There’s no world where it’s socially acceptable, or a good idea, to allow people to generate images based on publicly available Instagram profiles. Because, if I understand it correctly — I don’t use social media in my personal life — Instagram profiles are public by default. Influencers in particular thrive on putting as much content about themselves out there as they possibly can.
You have that interest clashing with Meta’s, which is to say: we have all of these images on our platform, so why don’t we use them to let people include them at inference for this new Muse model — by which I mean include them at the point you actually use it. Presumably they’re also included in the training data for Muse, because Muse seems to be a reasonably competitive image-generation model. But the part people are up in arms about is the permissionless side of things. Just like the REI story last week, which was based on Meta’s programmatic advertising stack, this is opt-out. So if you’re an influencer and you don’t want people generating images using your likeness, you have to go and manually set a flag to do that.
The model story is a bit more complicated, because it falls into the same zone where you have a retailer taking pre-existing material that they’ve presumably shot themselves. They’ve done a campaign with this guy before; they’ve done a PDP photoshoot with him before, and they’re treating that as fair game for intake. So at the point of input they’re presumably going: we’ve got some new products, we need to generate some new images, let’s use the guy we’ve used before — but let’s use him as a basis for iteration, and start to adjust his features, his skin tone, his height, all these kinds of things.
Now, there’s a sliding scale beyond which that person wouldn’t necessarily know he’d been used as the input. I’m assuming here that there are some giveaways that reveal he was the input, and that the output still superficially resembles him but with some transformations. You could do the same thing and transform somebody in a deep way — change something more profound about them — and they would probably never know.
The big issue this raises for brands and retailers is that you have to have a full chain of accountability, custody and visibility into how you generated your images. So if you’re using something like a node-based workspace, you need to know every step of that journey; you need to be able to reverse-engineer all of it to say what you did or didn’t do. Because, on the merits, this strays into some really uncomfortable areas around racism, and why anyone would want to adjust somebody’s skin tone to appeal to or appease a particular demographic — which is not great.
But it also comes down to this: if you’re going to use things as intake, you either need to sanitise your inputs and know that they’re above board, or you need full accountability — so that you can open your doors in these cases and say, look, what this person is claiming isn’t true; or, in the sketchier cases, what this person is claiming is potentially true.
AI’s physical footprint — data centres and community backlash
Grace Robinson: Like I said, AI really dominated the headlines this week, and the next story stays on the theme of the backlash against it. There was a really interesting piece in The Guardian about how the government has chosen Lanarkshire in Scotland as a key AI growth zone. The plan is to build large AI data centres, meant to help power all the AI models as AI becomes a bigger part of the economy. Unsurprisingly, local communities weren’t happy about it, but to me it was a stark reality check that AI really is having a physical impact on communities and on the environment.
So I wanted to get your take: as fashion adopts AI more and more, do you think it’s going to contribute to this problem? And how does fashion generally fit into this issue?
Ben Hanson: So I think some people seem very confused about where the backlash against AI is coming from. I’ve had conversations as recently as this week — I was a guest on another podcast, coming out at some indeterminate point in the future — and the host was saying, well, why do people hate AI so much? What makes it fundamentally different to 3D, or to the web, or to any other technology revolution?
There are two parts to it. One is the permissionless training piece we just talked about — the free-for-all image and text ingestion, everything else at the point of inference. There’s nothing to stop your creative work becoming part of a training dataset unless you simply don’t publish it on the internet. That’s one of the reasons people object. The other is that it’s having this tangible, physical effect on communities.
Now, data centres aren’t solely the preserve of AI, but AI is behind a lot of the data-centre build-out, because data-centre capacity and usage had been fairly static for a decade or so before this. Fashion isn’t responsible for all of that either. I’d argue fashion is probably no different from any other industry. Maybe in image generation — quantity and frequency — it’s higher than other sectors, but it’s going to be a negligible difference, I think. It’s a whole-society problem.
The issue with these is that data centres need a lot of space and a lot of power. Under the traditional cooling method — there are closed-loop cooling systems you can use that aren’t full of water, but most of them use open-loop water cooling — you’re basically taking a massive thing, putting it close to where people live, and that thing consumes a lot of power and resources. It can potentially raise utility rates in those areas, it’s an eyesore, and it’s apparently loud in a lot of instances.
But that’s almost the easy part, and almost not the worst part. You have data-centre build-out companies and energy companies promising these things will be beneficial. They’ll say: we’re going to amortise and absorb any rise in utility rates, so you won’t feel it; it’s going to create jobs; it’s going to charge up the local economy. What ends up happening is there’s very little job creation, and a whole lot of mess, noise and utility use.
This is part of a pattern that makes me sound like a crusty old left-leaning radical, which I’m not fully — which is that a lot of this objection to AI comes from the idea that the public and society must bear the cost, while private enterprise reaps the reward. You can frame it in pure anarchist terms as: socialise the losses and privatise the gains. This has that sort of sentiment to it.
As I said, fashion has hooks into this. It’s not solely responsible for it. But when people talk about the environmental and social impact of AI, these specific community issues are just as important as ‘I generate an image and it uses X amount of water or X amount of electricity’.
Digital wardrobes — Whering’s AI bet
Grace Robinson: The next story is a bit more positive — you’ll be happy to hear. This week there’s been a lot of coverage about Whering securing $7 million to ramp up its AI capabilities, from eBay and Google’s AI Futures Fund. For those who don’t know, Whering is a digital wardrobe app that helps people digitise their wardrobe, organise different outfits and really see what they have.
With the funding, they’re going to use AI to create personalised outfits based on the person’s mood, where they’re going, and even the weather. They’re also going to create more visually rich outputs, so it’s all levelling up. There have been a lot of digital wardrobe apps in the past, but now that Whering is incorporating AI, it feels like it might actually reach mass scale. So I wanted to get your take on what you think about this, and what the future holds for digital wardrobe apps like this.
Ben Hanson: Yeah. So this is one of those categories I recognise is growing, and compelling for a lot of people, but I don’t have a personal use case for myself, just because I try not to have too many clothes. There are some suits in my wardrobe I haven’t worn in ten years that probably shouldn’t be there. But aside from that, this is limited first-hand use for me. I did want to ask you, actually — is this something you’d use? Is it something you’d consider using? Do you use it?
Grace Robinson: Yeah, I think in the past I wouldn’t have used apps like this, but now that the quality and user experience have improved so much, I definitely would. And I think Whering seems to be the best example of it. I’d be open to it, for sure.
Ben Hanson: Yeah. And it dovetails with — so Google, interestingly, was the company that did that whole ‘we’re going to build the Clueless closet automatically from your Google Photos’ thing. They went through that whole process. Really smart tech, and again, a use case that’s not for me, but presumably a lot of people enjoyed it.
Now, these are the kinds of apps that feel like they’re B2C first. Save Your Wardrobe is another good example. They feel like consumer players, because in the first instance they are. What you’re trying to do is get people to download this and create behaviours, before you can build out the B2B, brand-facing and retailer-facing side of things.
I think AI will definitely help with the behavioural side, because the biggest barrier to somebody digitising their wardrobe is the effort involved — pulling every piece out, taking meticulous photographs from every angle, labelling everything, looking at the wash-care labels, checking the tags to figure out the material composition, all of that, guessing at the retail price if this is a resale play too. So anything you can do to reduce that friction helps create the user behaviour.
Now, I don’t know for sure that Whering has a B2B play to come here. Presumably it does, and the way I’d think of it is this. People talk about AI mediating between consumer and brand as part of an agentic product-discovery and shopping journey, and they talk about it passing the highest-intent, most complete customer over to a retailer. So rather than someone just rocking up and saying ‘I’m looking for X, help me,’ they say, well, I’ve already been through multiple rounds of queries, and not only am I looking for X, here’s a very finite definition of what X constitutes.
Add this to the mix and you can also say: actually, this pairs well with what you already have in your closet; or there’s an opportunity to upcycle something here; or to replace something here. If you tie this into take-back programmes and into AI commerce, all of a sudden it becomes much more interesting.
It’s the same philosophy behind how I feel about RFID: if you have item-level visibility — a tag that tells you where a piece of clothing is in the supply chain, in the store, in the fitting room and so on — that’s not that exciting on its own, but you can build some really cool things out of it. If what you start to get here is effectively the same thing — item-level visibility for what’s in people’s closets, what they have, what they’re looking for — then you can start to architect some really interesting things on top of it. So I agree: a more positive application of AI all round.
Sorting secondhand — AI for textiles and resale
Grace Robinson: So we’ve just spoken about AI helping consumers organise their wardrobes, but there were also some really interesting stories this week about AI helping organise secondhand fashion.
There’s a London-based startup called Fleek, which has just secured an additional $25 million in funding, and they’re going to use it to digitise the sorting of secondhand clothing. Fleek explains that so many garments go from donation bins to sorting centres, and the process of grading, pricing and organising them is still very offline, very analogue and very manual. So they’re going to use AI to completely revolutionise it.
At the same time, we saw a story from the Berlin startup reverse.fashion, which has also secured funding to develop AI-powered textile-sorting technology. Both of these companies are trying to replace labour-intensive, subjective sorting processes with AI, and this could potentially make the secondhand industry more sustainable and more efficient. So I wanted to get your take on this — is it a big leap forward, or what do you think?
Ben Hanson: So I think this is the kind of thing that can scale, depending on the hardware requirements, because it’s very easy to say you have AI that does this — it’s harder to know off the top of your head where that AI lives. The textile-sorting one, for example, feels like it’s part of the same class of technology as fabric quality-inspection AI, which has existed for a while using machine learning. There are now companies that do this for high-value fabrics. So if you think about high-quality leather, for instance, you want to be able to identify any defects in the hide before you cut it, so you can cut around them — and that increases the value of the final garment and reduces your wastage.
Now, that AI has to be tied to sensor hardware. You’re not just pointing your phone at a piece of leather; you’re actually putting it through a scanner to get those results. So how scalable, portable and durable this kind of setup is really depends on how much hardware infrastructure you need in place for it. Provided you can do that — and most of that hardware isn’t especially expensive — I think the AI sorting of textiles is fascinating, because it very quickly gets you to much more efficient recycling, upcycling and repurposing. And it’s not an untested or unproven avenue for AI: machine-learning AI has already been deployed in this space for a long time. So I’m optimistic about that side of things.
The sorting of secondhand goods is a similar deal. We’ve just talked about the importance of consumers digitising their own wardrobes, creating that behavioural pattern to enable brands and retailers to take it into first-hand sale. Now, if you think about the same thing — okay, I have an account with Vinted, I have an account somewhere else, whatever platform I use to sell garments when I’m done with them — if that platform can start to be more proactive about these things, then you can centralise it as well.
So you can say: give me a bag of all the things you don’t need. You don’t have to sort it. You don’t have to separate it into mono-materials and blends. You don’t have to separate it into tops and bottoms. You don’t have to determine what’s worth recycling and what isn’t. Send it to us, and we’ll do all of that automatically. We’ll centralise the whole process, and then presumably we’ll build partnerships that let us shift those things afterwards.
Optimistic about both. I think both of those have got some good runway in front of them.
Voice AI — OpenAI’s GPT-Live and the persuasion question
Grace Robinson: Now, to wrap up this week’s news, we’re ending with yet another AI story. This week, OpenAI announced that it’s going to upgrade its voice mode. The new GPT-Live experience is built on a full-duplex architecture, which means it can listen and speak at the same time to keep the conversation going. I’m pretty sure OpenAI said it’s meant to mimic a natural human conversation. It also introduces agentic capabilities, so the mode can delegate different tasks to other AI models while still keeping the conversation going. To me this feels equal parts exciting and concerning, but there’s no doubt it’s a massive leap forward — so I wanted to get your take on what it means for fashion.
Ben Hanson: Yeah, and I think my take on this one’s really quick and simple: it’s a big leap forward in voice interaction. OpenAI have always had the best voice mode, because if you’ve ever tried it in Claude or an equivalent, it works on a turn-by-turn, speech-to-text and text-to-speech basis. That introduces a lot of lag and a lot of inaccuracy. OpenAI have always had a really engaging, back-and-forth conversational dynamic, and swapping it to the duplex architecture just supercharges that. It’s probably going to be very similar to the sliding scales of expressivity you can build into the upcoming Siri AI, which is due out later this year along with iOS 27, I assume it’s called.
So that’s fine — the voice mode has been upgraded. The big part is that, as compelling as the voice mode used to be to interact with, it was always paired on the back end with one of their less capable, older LLMs. I think it was GPT-4.0 for a while; I’m sure it got upgraded at some point. But if you’ve seen any of those viral videos of the guy who talks to ChatGPT and says ‘set a timer for me, I’m going to run a mile,’ and then he comes back a second later — it has no concept of the passage of time. It’s a very interesting, human-seeming interface paired with a pretty dumb LLM on the back end.
So what they seem to have done is upgrade this to where it’s connected to the GPT-5.5 series, I think — or whatever, the 5.6, the planet Sol, Terra and Luna stuff. The interesting part for fashion is that this should then, at least in theory, upgrade it to where it’s connected to a lot of the brand-discovery, product-discovery and agentic-shopping stuff. So if you look at some of the video demos, they show weather search, World Cup scores, things like that — but it can present you with visual information.
So I don’t think there’s any impediment to companies like ASOS, that have done brand catalogues and immersive experiences, having that show up in voice interactions. That is a bit of a double-edged sword, though, because from a brand point of view, yes, it puts some compelling new surfaces out there. But it also really intensifies the question: as long as AI models and applications continue to be optimised for engagement — as long as they continue to be optimised for ‘can I do this for you, can I help you with this, can I do this with you’ — the more of that there is, the more chance there is of AI starting to actually railroad people into making purchases, starting to act in a salesman kind of role. And the most compelling persuasion tools human beings have at their disposal are your voice and your ability to generate pictures to visually represent things.
Adding a more compelling voice to this mix, I think, makes it much more dangerous, potentially. And to bring us full circle to what we talked about at the beginning, again it feels like an ask-forgiveness-later rather than ask-permission situation. I think we’ll see some things emerge from this — nothing quite as pronounced as the light AI-psychosis side of things, but I feel there’s a looming study, and I think we might have to do it if nobody else does. There’s a looming study about persuasion and prescriptive selling from AI still to be done. Let’s see how that pans out.
