“Virtual try-on” names two completely different technologies. One puts a live camera between the shopper and the product; the other never asks for a camera at all.
The short version: AR try-on and AI try-on get lumped together, but the line between them is the camera. AR overlays a product on a live camera feed and tracks it in real time (usually with a 3D model per SKU) — it answers “where does this sit in my space / on my face, right now?” AI try-on generates an image of a person wearing the item from a flat photo, no camera and no 3D model — it answers “what would I look like in this?” Cost, reach, and catalog coverage all flow from that one fork.
Augmented-reality try-on overlays a digital product onto a live camera feed and tracks it frame by frame — spatial registration to real-world geometry. It almost always needs a 3D model of each product and the shopper's camera permission. Unbeatable for true spatial fit (eyewear, furniture, watches to scale); the catch is it only runs for shoppers who grant the camera and tap in.
Generative AI try-on takes a flat photo of the shopper plus a product image and generates a new image of that person wearing the item — a diffusion model paints the garment on, preserving prints and proportions. No camera, no 3D asset; it runs from the product photo you already have. This is the family behind Google's Shopping try-on, which renders from one product photo across XXS–XXXL. The honest trade: it depicts rather than measures exact fit or drape.
One fork: AR registers an object to the real world in real time; AI generates a picture of a person wearing an item, after the fact. AR needs a live camera (and a 3D model); AI needs one static photo and your existing product image. AR is synchronous and answers “where does this sit in my world?”; AI is asynchronous, produces a shareable image, and answers “what would I look like in it?” The camera is the line between them — not the realism, and not the price.
The expectation is for confidence before buying, and neither technology owns it. 81% of Gen Z & Millennials expected AR to enhance their shopping (Klarna 2023, 5-country survey); 59% say a try-on of any kind helps them picture an item on themselves (Nosto); a 505,416-shopper meta-analysis found try-on raises purchase intent (Vieira et al., 2022).
AR's 3D-model-per-SKU and live-camera dependence scale well for a narrow, stable catalog (eyewear, watches, a furniture line) and poorly for a wide, fast-turning one. AI's flat-photo dependence scales to anything you can photograph — clothing, hats, bags, jewelry. They also reach different shares of traffic: AR only runs for shoppers who grant the camera and tap in (consumer AR engagement has been a persistent weak point); an AI image sits on the page for every visitor, and Google's AI try-on images earned 60% more high-quality views than standard photos.
A product page with 10,000 monthly visitors: an AI try-on renders for compute (~$0.067 each, our data), so even 10,000 try-ons cost ~$670 and every visitor can see a result on the page. An AR experience only runs for shoppers who grant the camera — if that's ~15%, you reach ~1,500 of 10,000 no matter how good the overlay looks. Same traffic; one architecture reaches several times more of it. That gap is the camera, not the quality.
Ask whether the value of seeing the product survives a still frame. If a still image captures it (does this dress suit me, does this hat work with my face) → AI try-on, because it reaches everyone and covers the whole catalog cheaply. If the value only exists live or in 3D space (do these frames sit right on my nose bridge, does this couch fit my room, is this ring the right size) → AR, because the value is the spatial registration. Many catalogs use AI broadly and reserve AR for the few SKUs where fit is everything.
AR wins on true spatial fit and scale, real-time interactivity, and live motion; AR try-on users were also 67% less likely to return and 80% more confident when they engaged (Snap + Publicis, N=4,028). AI wins on reach (on the page for everyone, no camera), breadth (any product with a flat photo), cost (no 3D pipeline), and speed to launch — its honest weak spots are exact measured fit and fabric drape.
If you sell clothing and mixed accessories and the job is confidence across the whole catalog, AI try-on is the pragmatic default — every visitor, every photographable SKU. That's the lane Ello is built for: 2D AI on the shopper's existing photo, no 3D models, no camera, covering clothing and accessories at ~$0.067 a try-on (our data). If true spatial fit is your whole product (eyewear, furniture), keep AR. See how the Shopify try-on apps line up, including the camera-AR one (Banuba) and the 3D/AR one (MirrAR), or look at real client results.
AR overlays a product on a live camera feed and tracks it in real time, usually with a 3D model — “where does this sit, right now?” AI generates an image of a person wearing the item from a flat photo, no camera or 3D model — “what would I look like in this?” The fork is the live camera.
Neither is universally better. AR is better for true spatial fit and scale (eyewear, furniture, watches); AI is better for reach and breadth (on the page for everyone, any photographable product). Pick by whether a still frame captures the value.
No. Generative AI try-on works from one static photo plus your product image — no camera permission, no per-SKU 3D asset. AR needs both a live camera and, usually, a 3D model per product.
If the value survives a still frame (most clothing, hats, bags, jewelry on the body) → AI, for reach and catalog coverage. If value only exists live or in 3D space (eyewear fit, furniture, ring sizing) → AR. Many catalogs use AI broadly and AR for the few fit-critical SKUs.