Virtual Try-On with AI: How Zara, Levi's and ASOS Are Reshaping Ecommerce Fit

The fitting room has always been retail's most expensive real estate — and its biggest conversion bottleneck. Online fashion has spent two decades trying to replicate the tactile certainty of trying on a garment, with mixed results. That changed meaningfully around 2022–2024, when major retailers began deploying AI-powered virtual try-on systems that go well beyond placing a flat image over a mannequin. Understanding what these systems actually do — and what they cannot do — matters enormously for anyone involved in garment construction, from independent designers to small ateliers.

What AI-Powered Virtual Try-On Actually Does

At its most basic level, virtual try-on maps a garment's 2D representation onto a human body model in a convincing way. The older approach, still common in budget implementations, overlays a product photo onto a static model image using simple affine transformations. The results look pasted-on and fail to simulate drape, tension, or how fabric behaves under movement.

Modern systems use a different pipeline. They combine computer vision for body segmentation, physics-based cloth simulation informed by real fabric properties, and neural rendering to produce photorealistic output. The consumer uploads a photo or uses a live camera feed; the system estimates body keypoints, infers a 3D body mesh from 2D signals, and then wraps the garment geometry over that mesh accounting for fabric weight, stretch, and construction seams.

What this requires on the garment side is precisely the kind of structured data that comes from pattern engineering: seam positions, ease allowances, grain lines, and fabric mechanical properties. A pattern that is well-documented and digitised produces better virtual try-on output than a pattern reconstructed from a scan of a physical sample. This is not incidental — it is the reason why the quality of the underlying pattern directly affects the quality of the virtual simulation.

How Zara, Levi's and ASOS Have Deployed It

Each of these retailers has taken a distinct approach, reflecting their production scale and customer base.

Zara introduced its virtual try-on feature — branded as a model try-on experience — in 2022, allowing customers to select a model that approximates their body type and height and see the garment on that silhouette. By 2023, Vogue Business reported that Inditex (Zara's parent company) had committed to producing digital product creation workflows for a significant portion of its collections, reducing the number of physical samples required before production sign-off. The system relies on garment digital twins created early in the product development cycle.

Levi's partnered with a digital avatar platform to offer personalised model diversity, specifically addressing the longstanding critique that fashion imagery fails to represent the range of bodies that actually buy clothing. Their implementation uses consumer-provided measurements to match against pre-rendered avatar bodies. The underlying premise is that fit confidence, not just aesthetics, drives purchase decisions — particularly for denim, where fit expectations are high and sizing inconsistency across the industry is well-documented.

ASOS has run its own version of fit technology through its Fit Assistant tool, which uses historical purchase and return data to generate size recommendations. More recently, ASOS has integrated visual try-on functionality that draws on the body data consumers already provide during the return process. According to reporting by Sourcing Journal, ASOS has cited a meaningful reduction in size-related returns in markets where the tool has been fully deployed, though the company has been cautious about publishing absolute figures.

The Returns Problem These Tools Are Solving

The commercial motivation is straightforward. According to data published by the National Retail Federation (NRF) in its 2023 Consumer Returns in the Retail Industry report, return rates for online apparel in the United States ran at approximately 24–26% of gross sales — versus around 8–10% for in-store purchases. A significant portion of those returns are attributed to fit and sizing issues. At scale, each returned parcel costs a retailer between €15 and €25 in logistics, restocking, and depreciation, before accounting for environmental cost.

A reduction of even 3–5 percentage points in the return rate for a retailer processing millions of orders annually translates into tens of millions of euros in recovered margin. That is the business case that justified the investment these companies made in virtual try-on infrastructure. The technology is not primarily a marketing tool — it is a supply chain efficiency play.

What the Technology Still Cannot Do Well

Honesty about current limitations is essential if this technology is to be evaluated seriously.

First, fabric texture and hand feel remain impossible to communicate digitally. A simulation can suggest how a fabric drapes but cannot convey whether a linen blend is crisp or soft, or how a jersey will pill after washing. This means virtual try-on is most effective for structural fit decisions (does this silhouette work on my body?) and least effective for material quality assessment.

Second, the accuracy of body mesh reconstruction from a single consumer photo is highly variable. Pose, lighting, background clutter, and clothing worn in the reference photo all introduce noise. Consumer-facing tools often compensate with generous tolerances, which can lead to systematically optimistic fit previews.

Third — and this matters most for independent creators and small ateliers — these systems require substantial data infrastructure. The major retailers can invest in building or licensing body scan datasets, training rendering models, and integrating digital twin workflows from early in the design process. For a designer producing 50 pieces per season, that pipeline does not exist off the shelf.

The gap between what enterprise retail can deploy and what is accessible to independent creators remains wide. A useful comparison:

What Independent Designers and Ateliers Can Actually Do Today

The practical takeaway for designers working outside enterprise retail is not to replicate what Zara or ASOS has built — that would require investment and engineering resources that are not realistic at small scale. The relevant lesson is upstream: the quality and structure of your pattern documentation determines how ready you are to participate in digital workflows as they become more accessible.

Garments built on well-drafted, size-graded patterns with documented ease allowances and construction logic are the input that any digital simulation system requires. Whether that simulation is used for virtual try-on, fit verification before cutting, or communicating construction intent to a remote machinist, the pattern is the foundation. A pattern that exists only as a physical template on brown paper, with measurements held in the maker's memory, cannot be fed into any downstream digital tool.

For ateliers and independent designers who want to move towards digital workflows without enterprise-level investment, the realistic starting point is digitising and structuring existing patterns, understanding grading logic, and working with tools that produce exportable, standards-compliant pattern files. MPattern is designed precisely for this entry point — enabling professional-grade pattern creation and management without requiring a CAD industrial background or a team of technicians.

The Fashion Institute of Technology and several European textile schools have begun integrating digital pattern workflows into their curriculum specifically because the industry consensus is that structured pattern data is the prerequisite for any subsequent digital application, virtual try-on included.

The Broader Industry Trajectory

Vogue Business and Business of Fashion have both tracked the acceleration of digital product creation across the supply chain since 2021, noting that the pandemic-forced compression of development timelines pushed brands to adopt digital sample approval workflows faster than previously planned. Virtual try-on is one visible consumer-facing output of a much larger shift toward treating garment data as a structured digital asset from the earliest point in the design process.

For the independent sector, this trajectory is both an opportunity and a pressure. Brands that have structured their pattern and product data well will find it far easier to integrate with wholesale platforms, direct-to-consumer ecommerce tools, and eventually virtual try-on systems as they become available at more accessible price points. Those that have not done that groundwork will face a catch-up cost.

The technology itself will continue to improve. Body reconstruction accuracy from single images is an active research area, with academic groups at ETH Zurich, MIT CSAIL, and elsewhere publishing advances regularly. Cloth physics simulation at interactive speeds has improved substantially in the past three years. The consumer experience of virtual try-on in 2027 will likely be materially better than today's. But the brands and designers who benefit most from that improvement will be the ones whose garment data is already structured and clean.

Conclusion

Virtual try-on with AI is solving a real commercial problem — ecommerce return rates driven by fit uncertainty — and the major retailers investing in it are doing so for margin reasons, not marketing ones. The technology works best when backed by rigorous pattern and garment data, which places pattern engineering back at the centre of any serious digital fashion workflow. For independent designers and small ateliers, the actionable priority is not to build a try-on system but to structure pattern documentation well enough that future integrations become possible. Explore how MPattern supports professional pattern workflows as a foundation for that transition.