Originally released in The Interline’s AI Report 2026, this executive interview with Vestir AI is one of a 17-part series that sees The Interline quiz executives from companies who have either introduced new AI solutions or added meaningful new AI capabilities into their existing platforms.
For more exclusive interviews, stories and opinions from different industry perspectives, and real survey feedback from voices from every level of fashion, download The AI Report 2026.
We’re observing a big disconnect (in both our own fashion-only data and in wider benchmarks from other sources) between the increasing capability and reliability of AI, in both a general sense and in discrete applications, and the lack of trust that end users place in its outputs. What does trust in your specific applications of AI look like, and how do you achieve and measure it?

I believe that anything new generates a natural sense of scepticism. The technology that underpins AI is moving so rapidly – every week we have new breakthroughs to incorporate and you really do need a research team to keep up! When things move at such a fast pace, it is only natural to feel uneasy. So although it seems paradoxical – increased capability and reliability being trusted less – it is no surprise that this is happening.

At Vestir, we have a full time research team that has been fully dedicated to solving the virtual try on challenge for 18 months. We have been very lucky to have a strong collaboration with ETH AI Center, part of ETH Zurich, one of the world’s leading universities. This has allowed us to really delve deep on the multiple challenges that prevent VTO from becoming more broadly adopted. 

Our solution is grounded in 3D, with a GenAI overlay. That makes our tech stack much deeper than many other offerings out there right now that rely on pure images to create “visualisations” as opposed to “try ons”. Right now, solutions that use photos as inputs face the challenge of their tech stack altering the shape of garments, and also bodies, to make the visualisation good. This creates something that looks engaging and exciting, but is not grounded in reality, which undermines trust when you receive your item at home and it looks nothing like the version of yourself you saw online.

Because of our 3D roots, we are able to provide accurate size recommendations where others fail, because we know the ground truth when it comes to precise measurements of both the avatar and the garment they are trying. And we have found a way to harness GenAI to stay faithful to that reality.

Our research collaboration has been funded by the Swiss Innovation Agency, which has stringent conditions on scientific measurement. We have had to regularly report on scientific accuracy, therefore we know our solution is accurate. Because of this, we are now able to provide size recommendations that people can trust, not just something that looks amazing.

Another element of trust relates to the sheer amount of data used to train AI models. People are rightly concerned that their data will be used inappropriately, or without their consent. At Vestir, we delete everything we can, where it is not needed. After people generate their avatars, all footage we have of the individual is removed. Remaining data is ringfenced on our platform, not shared with our B2B retail clients, which means it is all managed in one place and easy to delete if a user wants their data removed.

Ultimately, where there is rapid change, there is natural concern. It is the responsibility of AI companies not just to build robust solutions, but to act responsibly when it comes to data usage and put strong policies and procedures in place to help put their customers’ and users’ minds at ease. 

Virtual try-on, as a software category and as a consumer-facing experience isn’t new. In fact we’ve written before about how many times it’s been tried, with different technology foundations and different end user interfaces and demands. None of those previous attempts has really landed, but AI seems to have brought on a resurgence of excitement in VTO – and not just for investors, or the companies building it. But there’s a difference between excitement and results, so what do you see happening in the turn between VTO coming back as a frontend novelty, and it actually delivering on the outcomes it needs to?

The first incarnation of VTO was solely based on 3D computer vision. Whilst that could potentially solve the size & fit challenge, there were operational roadblocks when it came to garment digitisation, speed and scalability. The outputs also didn’t look that good – like garments had been “stuck on” to the background, inconsistency in lighting and plastic looking garment texture. This runs the risk of undermining brand image, so many brands & retailers didn’t want to take the risk.

Things still aren’t perfect, but the technology is moving so fast, that I believe it’s only a matter of time until these are no longer barriers.

But what has really changed in the past 6 to 9 months is the emergence of GenAI, and hence the ability to produce “try ons” that look like real video. In reality, up until now, the appearance has been a major blocker. Now GenAI can produce results that Retailers and Brands are finally comfortable putting on their website.

But behind the GenAI solutions is a major flaw. Because they are based on 2D photo inputs of both the customer and the garment, they have no basis of ground truth. By this I mean measurements. These solutions use diffusion models, which despite their training on thousands of pieces of data, have no education on the measurements of the person, or the measurements of the garment. They are also hard to control, which means they can change the shape of both the person and the garment to create an image that is not grounded in reality. As such, they can’t currently solve for size & fit. They look great – and may even improve conversion and engagement – but where is the utility?

To deliver across the outcomes virtual try on needs to and really add value, solutions need to provide utility as well as striking visuals. They need to be capable of showing a person how a garment is going to drape and move on their particular frame, as everybody’s body is different and unique. An effective virtual try on solution needs to show that. The ultimate goal is to be capable of providing a size recommendation, so that customers can trust that when they buy something, they will get it right first time, reducing returns for brands and retailers and frustration for customers. This is what we have been working on at Vestir – pairing the striking visuals with the reality of size & fit.

There’s obviously a heavy visual component to VTO. People want to see how a garment looks on them, and those aesthetics are likely to drive a lot of buying behaviour in the sense of converting more visitors to purchasers. But current buying behaviour also has the axiom of low-friction returns built in: customers are used to being able to buy something they like the look of, and then return it if it doesn’t fit. It feels as though virtual try-on has the potential to essentially fuel more of that wardrobing behaviour, if it increases conversions without measurably impacting returns. Do you believe the fit side of the equation has developed as quickly as the visual side?

As I mentioned before, now VTO is capable of reaching the realism bar, but the size & fit side is still lagging behind. It is easy to create a try on purely using GenAI – you can just plug a couple of photos into ChatGPT. 

But the only way to know whether a 3D garment fits on a 3D body is to have a 3D model of both. Creating a 3D human is relatively easy – now we can do that on a smartphone. But creating digital 3D garments is more of a challenge. Solutions such as Clo3D and Browzewear have come a long way, but these tools still lack widespread adoption as workflows and educating people how to use them take time and effort.

Right now creating 3D garments involves an array of different information – tech packs, sewing patterns, size charts… Some such as Zalando have volumetric capture facilities with set ups of over 100 cameras to take multiple images and reconstruct the garment from those. We were lucky to have access to one of these facilities as part of our collaboration with ETH Zurich, one of the world’s leading universities, but this sort of solution is not scalable.

There are many exciting fields of research right now that look to overcome the need for such complex processes, which our research team is leaning into heavily. Once these start to bear fruit, then the size & fit question can finally be solved at scale.

Following on from that previous question, there’s a difference between “solving” fit at the try-on stage, on a per-SKU basis, and addressing it more deeply. Brands and retailers are obviously trying to achieve greater fit consistency within their own walls, but there’s also no standardisation cross-brand and industry-wide. That feels like a challenge that VTO can only partially address, so what do you see as the more institutionalised blockers that are preventing return rates, as an industry-wide aggregate, from going down?

I think that’s exactly right – virtual try on is no silver bullet. It can help customers make better decisions at the point of purchase by reducing uncertainty: “How will this look on me?”, “Is this length right?”, “Does this silhouette suit my body?”. But it can’t single-handedly fix all size related returns. But the industry-wide blockers are deeper than the product page.

There are many solutions tackling the returns problem from multiple angles, and implementing a suite of different options can help remove some of the pain, but ultimately the root cause – inconsistency across brands, manufacturers and suppliers – needs to be addressed at source to truly solve the problem.

For sure there is an element of lack of discipline when it comes to standardisation across different manufacturers and quality control on the part of brands and retailers. When it comes to factory discipline, brands that have more control over their supply chain are better placed address this challenge. 

Strong process and practice can control / reduce manufacturer size variation, however it’s limited in its capacity to truly eliminate it. This is because variation also comes from fabric shrinkage, stretch recovery, cutting, sewing tension, washing and finishing. So there is always going to be a certain element of inconsistency.

One of the key challenges for any consumer-facing technology is that it needs to be available where the shopper already is, and it needs to meet them in a way that makes them feel included. How do you approach the dual challenge of there being so many different paths to purchase, across social channels and potentially AI-native ones, and there being relatively limited representation in traditional product photography?

VTO shouldn’t be treated as a feature simply buried on a PDP. Shoppers are discovering products on Instagram, TikTok, marketplaces, search, and increasingly through AI-assisted shopping journeys, so the experience has to be portable. Our ambition is to meet the shopper wherever intent is forming, and give them a clearer answer to two questions: “How will this look on my body?” and “Which size should I actually buy?”

The representation point is just as important. Traditional photography usually shows a narrow view: one or two models – “Anna is 5 ft 8 and wears a size 8”. That leaves shoppers doing a lot of mental work to imagine whether the item is right for them. At Vestir, our aim is not just to create a visual gimmick, but to give shoppers a more reliable indication of how a garment will look on their exact body, alongside a size recommendation they can actually trust.

The challenge is to do that without creating a new trust problem. The try on should not be overly beautified, reshaped, or made to behave in a way it would not in real life. That is why garment digitisation is such an important blocker: accurate virtual try-on depends on strong garment data, not just good-looking imagery. But as these new exciting research methods becomes commercially usable, the possibilities become much bigger.

Last year, we asked technology executives whether they predicted AI would become more obvious, as a primary interface paradigm, or whether it would disappear into the background in the way that vital but invisible platforms like AWS and Azure have done. This year’s data suggests that the answer is both: fashion professionals that use AI interact with it as both the engine and the steering wheel. But we also see that, for a lot of companies, both kinds of applications are still largely in the scoping or refined-pilot stage. What do you see being the trigger for deeper adoption and wider roll-out?

We have to remember that AI is still really in its infancy. It takes time to test anything new. Larger companies are also faced with multiple challenges that simultaneously compete for attention. There may be many internally competing projects or bureaucracy that stands in the way of orderly progress. It’s also impossible to find one silver bullet that will solve all your woes. So finding that right combination of solutions that move the needle in the most meaningful way is a significant challenge that takes time.

But in my opinion, the major trigger to broad scale adoption of new technologies is the ability to provide actual utility. This counts for both the end consumer as well as retailers and brands. Without utility, gimmicks are fun for a period of time, but when the novelty wears off, what is left? VTO may look amazing with what GenAI can produce today, and it may drive more sales, but without size recommendation, it is broadly useless. 

From a customer standpoint, people will still have to return stuff. From the retailer standpoint, your bottom line still suffers. Until any technological solution can provide utility, it has no value.

There’s also the opportunity and ability to measure and prove value. Nobody wants to be the first and take a plunge until it is proven elsewhere, but then you’re in a chicken egg situation. Once you do take the plunge, how do you isolate the impact in a meaningful way? How do you support roll out in the best way, so you actually prove it on a large enough scale to make a meaningful difference? Until we take braver decisions, it’s hard to prove anything.

You May Also Like
All About AI: Juan Pellerano of Swap Commerce
Read More: All About AI: Juan Pellerano of Swap Commerce

All About AI: Juan Pellerano of Swap Commerce

Industry data suggests that fashion professionals and shoppers distrust AI. Real-world behaviour indicates the opposite, and agentic commerce is being built on that latter bet.
All About AI: Maísa Benatti of AIUTA
Read More: All About AI: Maísa Benatti of AIUTA

All About AI: Maísa Benatti of AIUTA

What it means to build trust in AI Virtual-Try On, for that trust to hold up in large-scale deployments online and off, and for brands to bring genuine fashion expertise to bear on AI outputs.
All About AI: Matt Rouif of Photoroom
Read More: All About AI: Matt Rouif of Photoroom

All About AI: Matt Rouif of Photoroom

To truly become the next stage of fashion's imaging pipeline, generative models and tools need to be able to represent and honour the product promise, and deliver a return that goes beyond the token level.