Where Specialist Virtual Try-On Systems Claim An Edge

The specialist-AI case becomes even more pertinent when you look at what virtual try-on still finds difficult.

A 2024 survey from Brunel University on deep-learning virtual try-on models points to persistent weaknesses around preserving clothing characteristics and textures, accurately applying clothing to the person, preserving facial identities and dealing with dataset bias. Deloitte has made a similar point from a retail perspective, noting that in fashion “accurate fit and complex features, like the drape of varied fabrics, can be particularly difficult to render effectively right now.”

This is precisely where DressX says its approach differs, training on large-scale fashion-specific datasets and video-based inputs, allowing it to analyse garments in motion through computer vision and capture “drape, structure and fabric dynamics” more precisely than static, image-based systems. It also says the system incorporates styling logic and brand-specific attributes so that garments are shown in a way that reflects each brand’s visual identity rather than producing generic outputs.

Modenova argues that motion data is the missing piece. “Our video-based training layer allows us to capture how garments behave in motion, not just how they look in a static image. A general system can render something that looks plausible, but fashion brands need it to be correct. That means how a jacket sits on the shoulders, how fabric falls when someone moves, or how a silhouette holds its shape. Motion data is key here. It allows the system to understand how materials act in real life, which is what ultimately defines whether the result feels accurate from a brand perspective."

While 3D-based virtual try-on relies on simulated physics and avatar-driven approximations, DressX’s approach is more rooted in reality, trained on real-world garment behavior across diverse bodies, with "a precision that elevates digital try-on from approximation to true representation, a standard essential for luxury.”

There is a temptation, especially in fashion tech, to treat these specialist claims as proof that large horizontal players will struggle to matter, although the reality appears to be more layered than that. Google is already using a fashion-specific model, and its advantage in discovery and distribution is enormous. What DRESSX is arguing for is a different kind of value: the point where brand-correct outputs, premium presentation and category-specific nuance start to matter more than overall reach.

From Virtual Try-On Feature To Merchandising Layer

DressX’s overview makes this ambition explicit in their overall philosophy. It says the platform has generated more than 23,000 virtual try-ons to date, and most recently partnered with Victoria Beckham to introduce AI-powered try-on directly within the brand’s ecommerce environment.

Shapovalova says the category is already moving beyond novelty. “Try-on is no longer just a consumer-facing visual trick. It’s becoming part of the infrastructure behind how fashion is marketed and sold. As this behavior becomes native, brands increasingly begin to optimize their merchandising, content, and distribution strategies around try-on as a core touchpoint in the customer journey.”

She adds that the company is already seeing “some exciting data that confirms that try-on users have 10x conversion to purchase versus users that did not do the try-on on the same PDP. In addition to conversion, it’s also driving engagement: users who engage with try-on even once demonstrate up to 7x higher retention compared to those who never interact with it, reinforcing its role not just in conversion, but in long-term customer engagement.”

This broader infrastructure story is also an area of focus for NVIDIA. In January 2025, Nvidia launched a retail shopping assistant blueprint with virtual try-on built directly into an online chat interface; part of a system designed to drive higher conversion rates, lower product return rates and increase order size through more personalized suggestions. In other words, another large technology player is also treating try-on less as a standalone experience and more as a layer inside a wider retail workflow.

Google is doing something similar from the consumer side. Its AI Mode shopping experience ties try-on to search, product discovery, agentic checkout and the Shopping Graph’s 50 billion listings.

Seen through that lens, the market is starting to break into three layers: Google is normalizing try-on for the mass shopper, Nvidia is helping build the retail rails around AI-assisted discovery, and DressX is trying to own the specialist fashion layer where styling, merchandising and brand fidelity matter most.

What Still Is Not Solved For Virtual Try-On

This specialist thesis becomes more convincing when try-on’s current limits are clearly defined. Modenova does that better than many founders in the category. “AI try-on still has its challenges, but the progress over the past few years has been significant. Areas that were previously very difficult, like complex layering or styling multiple items in one look, can now be represented much more accurately than before. At the same time, fashion remains a highly fragmented industry, where sizing, measurements, and garment specifications are often inconsistent, not only across brands but even within the same brand across different collections or production runs. This lack of standardization is one of the core underlying challenges.”

She continues: “The quality of the output still depends on the input. Consistent product imagery, sizing data and garment metadata remain essential to achieving the best results. In many cases, brands are missing key inputs entirely, whether it is precise measurements, complete metadata, or even sufficient imagery.”

One of the key recurring problems in AI retail is the distance between a compelling demo and a usable system. Beyond the need for better rendering, fashion brands also need cleaner data, more disciplined workflows and a realistic understanding of what try-on can and cannot promise.

The confidence numbers are central to this argument. DressX says its virtual try-on can give customers up to 89% confidence in fit. According to Modenova, that figure is “based on post-interaction surveys and behavioral data, benchmarked against users browsing the same PDPs without try-on.” She says the “strongest impact appears at the confidence stage, driving add-to-cart and conversion,” especially when shoppers can move from single-item try-on into full-look and mix-and-match combinations.

While being persuasive, this also provides a reminder that visual confidence and true fit confidence are not identical. The category still has to prove, repeatedly and at scale, that better imagery translates into fewer returns, better customer satisfaction and more trustworthy digital shopping.

The Layered Future Of Fashion AI Virtual Try-On

So, which approach wins out in the end? The most credible outcome for this market is coexistence.

Shapovalova is explicit about that. “The most effective approach is likely a combination of both. Large platforms provide scale, while specialized AI delivers the level of accuracy and nuance that fashion requires.”

Modenova’s version is more commercially pointed: “General-purpose platforms might win in general distribution numbers, but specialist fashion AI layers will win in a specific distribution niche that actually is a more profitable cohort, and a more predictable revenue driver.”

That also helps explain where DressX wants to go next. Shapovalova says virtual try-on is only “one part of a much larger system,” with the next layer being “how these technologies connect across the full lifecycle, from content creation and campaign production to merchandising, personalization and sales.” Modenova points to AI agents and more concierge-like shopping experiences, where try-on sits inside a broader layer that actively drives discovery, styling and conversion.

Rather than being a comparison of ‘David versus Goliath’ approaches to the same issue, virtual try-on seems to be moving more towards a vertical specialization versus horizontal scale landscape, where consumer confidence can be increased on the path from product page to purchase, with a focus on brand-correct garment representations over plausible ones.

Google has made virtual try-on feel natural at scale, DressX is making the case that fashion still needs a specialist layer above that scale, and Nvidia’s entry suggests the infrastructure story is also becoming more important. The next phase of fashion AI looks more likely to be a stack in which each layer solves a different problem, rather than a winner-takes-all scenario. But the real question now is which of those layers ends up owning the most value for brands.