Eyewear is the category that made virtual try-on famous, and AR is very good at it. But there's a fork hiding inside "eyewear," and sunglasses fall on the side most people miss.
The short version: AR eyewear try-on works like a virtual mirror — it maps a frame onto your face through the camera in real time — but it only works for frames someone has digitized into a 3D model first. The largest library has built 195,000+ of them, one per SKU (Fittingbox). For sunglasses that precision is aimed at the decision shoppers make least: sunglasses are chosen on face shape and style, which a still image answers. 2D AI renders any frame from a flat product photo — every SKU, every new drop, seen by every shopper, at ~$0.067 each. AR still wins where fit is the point (prescription eyewear); for sunglasses, 2D AI covers more catalog for less.
Because AR eyewear try-on is a virtual mirror: it places a digital frame on your face via the front camera and tracks it in real time. The catch is what it needs to run — a 3D model of every single frame. That's the whole industry: Fittingbox, the 15+ year pioneer, has 195,000+ frames from 1,200+ brands digitized in 3D and runs 215 million virtual fittings a year. The library exists precisely because AR can't overlay a frame it hasn't modeled first.
For prescription eyewear, AR's precision earns its keep — where the frame sits on your nose bridge, how lenses align with your pupils, arm length. Sunglasses are a different job: almost no one buys them on lens-fit precision; they buy on whether the shape flatters their face. Every eyewear retailer's own advice confirms it — Ray-Ban, Oakley, Warby Parker all sell sunglasses through face-shape guides, not fit calculators. A face-shape judgment is exactly what a single clear image answers.
2D AI skips the 3D library entirely: it generates an image of the shopper wearing the frames from your existing flat product photo — no camera, no per-frame 3D model. This is the family behind Google's Shopping try-on, whose AI try-on images earned 60% more high-quality views than standard photos. AR can only try on frames modeled in 3D — fine for major brands already in a library, but your own-design or fast-turning styles aren't in one, so each new drop waits on digitization. 2D AI renders any frame you can photograph.
| AR overlay (camera) | 2D AI try-on | |
|---|---|---|
| Input | Live camera + a 3D model of each frame | One flat product photo you already shot |
| Catalog coverage | Only frames modeled in 3D (a library, or newly digitized) | Any frame you can photograph — new drops day one |
| Who sees it | Only shoppers who grant the camera | Every shopper — it sits on the page |
| Best at | Real-time fit: nose bridge, pupil alignment, arm length (prescription) | "Does this shape suit my face?" — the style call |
| Honest weakness | Per-SKU 3D cost; most shoppers never open the camera | Depicts style; doesn't measure precise lens fit |
An independent brand with 120 sunglasses SKUs (frames × colorways) that drops new styles every season. For AR, each frame needs its own 3D model before anyone can try it, and your own designs aren't in a pre-built library, so every drop waits on digitization. For 2D AI: $0 asset prep (it uses your photos) and at ~$0.067 a try-on, 8,000 try-ons a month costs about $536 — every frame, every shopper, the day it launches. On reach: a 2D AI image sits on the page so 100% of shoppers see a result, whereas AR only runs for the minority who grant the camera.
For "do these suit me enough to buy?", yes — the lift comes from confidence, not geometry. Try-on users were 80% more confident and 67% less likely to return (Snap + Publicis, N=4,028); a 505,416-shopper meta-analysis found try-on raises purchase intent (Vieira et al., 2022); 59% say a try-on helps them picture the item (Nosto); and 81% of younger shoppers already expect some form of try-on (Klarna, 2023). None of that needs a 3D model.
When fit is the decision (prescription eyewear — lens position and how the frame sits), when live motion matters (do frames slip when you look down), and when you're a big multi-brand retailer whose SKUs are already digitized in a library. In those cases AR earns its per-SKU 3D cost. The argument here is narrower: for a brand selling its own sunglasses on style, 2D AI covers more of the catalog for a fraction of the cost.
Three checks: are your frames already in a 3D library (if they're your own designs, almost certainly not); are shoppers deciding style or prescription fit (sunglasses are style); and how fast does your catalog turn. If the answers point to style and speed, 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 including glasses, at ~$0.067 a try-on (our data). Compare the Shopify try-on apps, including the camera-AR one (Banuba) and the 3D/AR one (MirrAR), or see real client results.
Yes. A 2D AI try-on generates the sunglasses on the shopper's own photo from your existing product image — no camera and no per-frame 3D model. AR overlay try-on is the approach that needs a 3D model of every frame, which is why libraries like Fittingbox have digitized 195,000+ of them.
It depends on the decision. Sunglasses are chosen on face shape and style, which a still image answers well, so 2D AI is usually the better fit — it covers your whole catalog and every shopper sees it. AR is better for prescription eyewear, where real-time lens and nose-bridge fit genuinely matter.
Because AR overlays the frame on a live camera feed and tracks it on your face in real time, which requires 3D geometry for each SKU. That per-frame requirement is why the largest provider has built a library of 195,000+ frames in 3D. 2D AI skips it by generating an image from a flat product photo.
It shows style and how a frame suits your face very well — which is what sunglasses buyers decide on. It depicts rather than measures precise lens positioning, so for prescription eyewear an AR or measurement-based tool is stronger. For sunglasses, that limit rarely bites.