Virtual Try-On for Jewelry: What Actually Works in 2026
Short answer: Rings are the hardest virtual try-on category in retail, harder than glasses, watches or makeup, for reasons that are geometric rather than commercial. Good implementations exist and they help. Most implementations are demos that customers try once, and the money would have been better spent on scale references and a real size guide.
This post is about which of those two you are about to build.
The problem the numbers describe
Jewelry converts worse than almost anything else sold online. Industry benchmarks put jewelry ecommerce conversion around 1.5 percent against a cross industry average of about 2.3 percent, mobile conversion around 1.6 percent against 2.85 percent, cart abandonment near 74.8 percent, and returns around 20 percent against a 17.5 percent average.
Read those together and the diagnosis is not mysterious. People arrive, they cannot resolve their uncertainty about an expensive object they cannot hold, and they leave. The ones who do buy send a fifth of it back, and the reasons are overwhelmingly about scale and proportion rather than taste. It looked bigger. It looked smaller. It sits differently than I expected.
Virtual try-on is aimed squarely at that gap, which is why it is attractive, and why it is worth being precise about what it does and does not close.
Why rings are the hard case
Face based try-on, for glasses and makeup, sits on a decade of mature, well funded face tracking. Faces are large, high contrast, roughly rigid, and the models that find them are excellent and freely available.
Hands are none of those things.
The tracking is weaker. The available open source hand tracking is good at locating knuckles and estimating overall hand size. Placing a ring convincingly needs more than that: an estimate of the width of a specific finger, stable landmark detection through motion, and a stack of approximations on top. There is no equivalent of the off the shelf face mesh for fingers, which is precisely why the good ring try-ons are proprietary and the quick ones look wrong.
Occlusion is unforgiving. A ring wraps a finger. Part of it must be hidden behind the finger, and the boundary between hidden and visible moves as the hand moves. Get it slightly wrong and the ring appears to float, and the illusion does not partially fail, it fails completely.
Scale is the entire point and the hardest thing to get right. For glasses, approximate scale is fine, because the customer is judging style. For a ring, scale is what the customer came to check. A try-on that renders every ring at a plausible but uncalibrated size is answering the wrong question convincingly, which is worse than not answering it.
Materials are judged by an expert. Your customer has been looking at ring photographs for months. They know what platinum looks like against skin and how a brilliant cut throws light. Real time rendering on a phone, over a camera feed, with whatever lighting the room provides, is being asked to satisfy a trained eye at the exact moment they are deciding to spend a lot of money.
On the web, all of this is harder. Browser based AR has to work across Safari, Chrome, Firefox and the rest, negotiate camera permission, run photoreal rendering inside a browser's resource limits, and stream assets over whatever connection the customer has. A native app avoids most of that and reaches far fewer people. That tradeoff has no clean answer, and anyone who tells you otherwise is selling one side of it.
What genuinely works
Calibrated scale, not just placement. The implementations that reduce returns are the ones that solve for real size, using a reference object or an established finger size rather than rendering everything at a comfortable default. This is less impressive in a demo and far more useful in a sale.
Stills over live video. A photograph the customer uploads or captures, processed once with time to do it properly, beats a shaky live overlay on almost every axis: better lighting, better occlusion, better materials, no jitter, and it works on any device. It is also shareable, which live AR is not, and sharing is a real part of how engagement rings get decided.
The pair, not the piece. For wedding bands the question is almost never "how does this band look" and almost always "how does this band look next to the ring I already wear". A try-on that can show the pair is answering the actual question. Most cannot.
Try-on as one part of a visual system, not a feature. The same 3D models that drive a configurator drive the try-on, the product page, the in store screen and the video. Built that way the try-on is close to free, because the expensive part, the model library, already exists. Commissioned as a standalone feature it is expensive and it stays expensive.
What usually does not
A hand from a stock library. Displaying a ring on a model's hand is a product photograph with extra steps. It tells the customer nothing about their hand and it is not try-on, whatever the button says.
Uncalibrated live AR. Impressive for about eight seconds. Used once. It generates a spike of engagement metrics that look like success and no measurable change in conversion or returns, because it never addressed the uncertainty that was stopping the purchase.
Try-on before the fundamentals. If your product page does not yet have a proper scale reference, real dimensions in millimeters, a coin or finger comparison, a clear size guide and a return policy that removes the fear, then those are cheaper, they work on every device, they help every visitor rather than the fraction who grant camera access, and they address more of the returns than AR does.
An honest ordering
If the goal is fewer returns and more completed purchases, spend in this order.
First, accurate dimensions and scale references on every product page. Cheapest, works everywhere, addresses the largest single cause of returns.
Second, real time 3D on the product page. The customer rotates and inspects the actual piece at true proportions. No camera permission, no tracking, works on a mid range phone if the assets are budgeted properly, and it uses the same model library everything else will need.
Third, still image try-on. Good results, broad device support, shareable.
Fourth, live AR try-on, if your category and audience genuinely want it and you are prepared to do calibration properly rather than shipping a placement demo.
Most jewelry businesses are being sold the fourth thing while the first is still missing.
What we do
At AMG Dynamics we build the model library first and treat try-on as one output of it rather than a standalone feature, because that is the only version where the economics work. We are also willing to tell a client that scale references and a better size guide will do more for their return rate this quarter than AR will, which is often true.
We work with jewelry brands, manufacturers and retailers from London, Delaware and Mumbai, across 3D configurators, visualization, workflow automation and AI systems.
Want a view on whether try-on is the right next spend for your catalog? Send your product page URL and your current return rate to admin@amgdynamics.com.
The model library that makes all of this affordable is on our 3D jewelry visualization page.
Frequently asked questions
Does virtual try-on work for jewelry? It works when it solves for true scale rather than just placing a ring on a hand. Rings are the hardest AR category because hand tracking is weaker than face tracking, occlusion around the finger is unforgiving, and scale is the exact thing the customer is trying to judge. Calibrated implementations help with returns. Uncalibrated ones produce engagement without changing conversion.
Why is ring try-on harder than glasses try-on? Face tracking is mature and freely available, and approximate scale is acceptable because customers are judging style. For rings there is no equivalent off the shelf finger tracking, the ring must be correctly occluded by the finger as the hand moves, and scale is the whole question rather than a detail.
Is web based AR good enough, or do I need an app? Web reaches everyone and constrains rendering quality, device support and asset loading. Native apps give better tracking and rendering to the small fraction of customers who install them. For most jewelry retailers the better answer is still image try-on on the web, which sidesteps most of the tradeoff.
Will virtual try-on reduce my return rate? Only if it renders at true scale. Jewelry returns run around 20 percent and are driven mostly by size and proportion surprises rather than taste. A try-on that shows a plausible but uncalibrated size does not address that, while accurate dimensions and scale references on the product page do, at a fraction of the cost.
What should I build before try-on? Accurate millimeter dimensions and a physical scale reference on every product page, then real time 3D that customers can rotate and inspect. Both work on every device without camera permission, both use assets you will need anyway, and both address more of the uncertainty than a try-on demo does.