1990s: mall kiosk experiments
The earliest documented virtual dressing room concepts appeared in retail research labs and academic papers in the early 1990s. The implementation typically involved a camera, a mirror-shaped display, and rule-based graphics overlays that could superimpose a simplified clothing silhouette on a video feed. IBM and several European retail groups ran limited pilot installations in department stores and mall kiosks between 1994 and 1999.
These systems were impractical for two reasons: the compute required for real-time video processing was expensive and physically large, and the graphics quality was far below the threshold where shoppers found the result believable. Adoption was uniformly low. The projects were useful as demonstrations that the concept could work in principle, but the technology was decades away from commercial viability.
2010s: the AR and mobile camera moment
The proliferation of smartphones with front-facing cameras and dedicated image signal processors created the first viable mass-market AR try-on platform. Snapchat's Lens Studio, launched in 2017, democratized face-AR creation and demonstrated that tens of millions of users would engage with real-time augmented reality experiences when the latency was low enough and the render quality was high enough.
Fashion and beauty brands moved quickly into this space. Sunglasses brands built lens-style try-ons for glasses. Cosmetics brands offered real-time lip color and foundation shade previews via AR. These applications worked well because they required tracking a relatively rigid surface — the face — which is a more tractable problem than tracking draped fabric on a moving body. By 2019, AR try-on was a proven commercial tool for accessories and beauty, but still largely unproven for apparel.
2018–2020: first-generation e-commerce try-on
The first wave of e-commerce virtual try-on products — targeting apparel specifically — launched between 2018 and 2020. These products typically used a combination of body pose estimation (estimating the 3D position of body joints from a 2D image) and texture mapping to drape a 2D garment texture onto a detected body silhouette. The results were technically impressive but visually unconvincing: fabric edges were poorly defined, lighting was inconsistent, and complex garments like layered outerwear or flowing dresses produced artifacts.
Commercial adoption was limited. Several well-funded startups in this space either pivoted to B2B catalog photography or closed between 2020 and 2022. The fundamental problem was not compute or engineering effort — substantial capital was deployed on both — but model architecture: texture-mapping approaches could not realistically simulate how fabric drapes, folds, and interacts with body geometry.
2022: the generative AI inflection
The release of latent diffusion models with sufficient resolution and control mechanisms — the technical foundation of image generation systems that emerged prominently in 2022 — changed what was possible for virtual try-on in a fundamental way. Instead of texture-mapping a garment onto a body, diffusion-based models could generate a photorealistic image of a person wearing a garment, conditioned on both the person's photo and the garment's appearance. The fabric drape, lighting interaction, and body occlusion all emerged from the generation process rather than from explicit simulation.
This architectural shift is what made photo-based apparel try-on a commercial product. Photta launched its B2B widget powered by Nano Banana 2, a fine-tuned diffusion model optimized for fashion and jewelry applications, as part of this generative AI era. The render quality crossed the threshold that drives actual commercial outcomes: shoppers found the results believable enough to make purchase decisions based on them, as evidenced by the conversion and return-rate data from Photta's merchant cohort.
2026: where the technology stands
As of 2026, generative AI-based virtual try-on is a mature commercial product for apparel and jewelry. The technology delivers consistent photorealistic results at acceptable latency (8–15 seconds), scales to catalogs of any size without per-SKU production overhead, and has accumulated enough merchant deployment data to support reliable ROI benchmarks. The question for a fashion merchant in 2026 is not 'does this technology work?' but 'which implementation fits my catalog and traffic level?'
Adjacent applications remain earlier in their development cycle. Footwear try-on presents specific challenges around foot geometry and sole rendering that apparel diffusion models do not address well. Video-format try-on — generating a short clip rather than a static image — is in active development but not yet at the render quality threshold for commercial deployment at scale. Multi-garment outfit composition (try on a top, bottom, and accessory simultaneously) is an active research area with early commercial implementations beginning to appear in 2025–2026.