Does the customer have to install an app?

For a filter on Snapchat or TikTok they are already inside that app, which is the point of publishing there. For try-on on your own site, no install is required at all — it runs in the browser, asks for camera permission, and works from a link you can put in an email, an ad or a product page. Removing the install step is usually the largest single gain in how many people actually try it.

Do our customer photos get uploaded anywhere?

No. The camera frames are processed inside the visitor's browser and nothing is sent to a server, which also means no image storage, no retention question and a much shorter conversation with your legal team. We state that plainly on the page itself, because people hesitate before granting camera access, and what they hesitate over is where the pictures go — so the page answers that before it asks.

How accurate is the sizing on different faces?

The model is not placed at a fixed size. Position and rotation come from the line between the eyes and the scale from the bridge of the nose, so the frame resizes itself to each face rather than assuming an average one. It is convincing rather than optical-grade, and it is meant to answer does this suit me, not what is my exact pupillary distance.

How many products can go into a try-on?

As many as you like, but each one is real work — its own model, its own materials and its own scale calibration against the face landmarks. Product count is the biggest single driver of the price. Starting with a focused set of bestsellers keeps the first build small enough to judge on its own conversion figures, and what those show tells you which products are worth adding next.

What happens on a cheap phone or in bad light?

Tracking degrades before it fails — the overlay drifts or the frame rate drops — so there is always a designed fallback rather than a broken screen. Usually that is a still-image mode or the standard product gallery, offered automatically with a clear message. The page still has to sell to the visitor whose phone cannot run the clever version.

Should we build on one platform and stay there?

No, and this is a hard-won point. Meta shut its Spark AR platform down in 2025, and every effect and every workflow tied exclusively to it went with it. Effects are built so the underlying assets and the concept are portable, and anything that has to live long-term belongs on your own site rather than inside someone else's tooling.

Immersive, 3D & Emerging Tech

AR try-on and social filters that hold on every face, not just yours

Snapchat and TikTok effects, or web-based try-on that runs in the browser with no app install — tracked, smoothed, scaled per face and tested on faces that are not the team's.

From: $1,500–$20,000
Typical time: 3–10 weeks
Usually bought by: Beauty, eyewear, fashion

What this actually solves

Two quite different problems arrive under the same heading, and it helps to separate them before deciding what to buy.

The first is reach. A brand launching a product, a film or a season wants something people pass to each other unprompted, and a filter is one of the few marketing assets whose entire mechanism is that the audience distributes it for you. Nobody shares a banner ad. They will happily put a crown, a pair of frames or a team's colours on their own face and post it, and every one of those posts carries your product into a feed you did not pay for.

The second is a sales problem, and for some categories it is the whole business. Eyewear online lives or dies on one hesitation — the customer cannot picture the frame on their own face, so they do not buy at all, or they order several and send back the ones that turned out not to suit them. Cosmetics has the same gap between a swatch on a website and a shade on a person. Every return is a shipping cost, a restocking cost and a customer who now doubts the next purchase too. Try-on removes the hesitation at the exact moment it occurs, on the product page, without asking anyone to download anything.

The two are worth buying for different reasons and should be judged by different numbers. A filter is measured in reach and shares. A try-on is measured in conversion on the products that have it and in the return rate afterwards. Confusing the two is the most common reason this work disappoints — a beautiful filter judged on sales, or a try-on judged on how many people posted it.

How the work runs

Decide the surface, and confirm the products. Whether this lives inside Snapchat and TikTok, on your own site, or both, and exactly which products are in scope. Each product means its own model and its own calibration, so this decision sets most of the cost. We also check what 3D assets exist — eyewear brands often have supplier CAD, and beauty rarely has anything. Two to four days.
Prepare the assets. Modelling or cleaning the product, keeping it deliberately light because it will be composited over a live camera feed on a phone that is also running face detection. Materials tuned for a moving, uncontrolled lighting environment rather than for a studio render, and the anchor points defined — where on the model the nose bridge sits, where the temples fall. One to three weeks.
Build the tracking layer. For web try-on, face landmarks are detected in the browser from the camera stream, targeting a smooth frame rate on mid-range hardware, and the product is positioned from the eye line and scaled from the nose bridge so it fits each face rather than an average one. Raw landmarks jitter, so a smoothing filter is applied and tuned — too little and the frames vibrate, too much and they lag behind the head. Scene lighting is sampled from the camera image so the product does not look pasted on. For platform filters, the equivalent work happens on the face mesh inside Lens Studio or Effect House. One to three weeks.
Test across real faces, and on weak devices. This is the phase that separates a demo from a product, and it cannot be shortened. At least ten genuinely different faces — beards, glasses already worn, different face widths, ages and skin tones — plus poor lighting, backlighting and a low-end Android. Every failure found here is a calibration or fallback change, not a rebuild, provided it is found here. One to two weeks.
Publish, or integrate. Platform effects go through Snap or TikTok review, which has its own queue and its own rejection reasons, and are published from your brand account rather than ours so they stay yours. Web try-on is embedded into the product page with camera permission handled gracefully, a photo capture button, the privacy statement, and the fallback path wired in. Three days to two weeks.

What moves the price

The range is $1,500–$20,000 across 3–10 weeks, and the spread is genuinely this wide because two different products share the page.

Product count. A single hero frame and a forty-piece eyewear collection are not comparable. Each additional product carries modelling, materials and its own scale calibration against the landmarks, and that is the line that grows fastest.

A shareable effect versus a commerce feature. A social filter is a self-contained asset with a review process at the end. A try-on wired into a live product page inherits your analytics, your privacy statements, your device mix and your uptime expectations, and sits at the top end of the range.

Whether a paid SDK is required. Browser-based face tracking can be done with free, well-maintained tooling. Certain effects — surface tracking, world AR, particular platform features — need a commercial SDK with its own subscription, which is a recurring cost of yours rather than a one-off of ours, and we say so before it is committed.

What usually goes wrong

It gets tested on the face of the person who built it. The frames sit perfectly on the developer and slide off a rounder face, float above a beard, or collide with the glasses somebody is already wearing. This is the defining failure of try-on work and it is invisible until a real audience arrives, at which point the fix is public. Ten different faces before launch is the cheapest insurance in this service.

The whole thing is anchored to one platform. Meta closed Spark in 2025 and the effects and workflows built exclusively inside it stopped being assets. Any AR capability that matters commercially should have its centre of gravity on your own site, with platform effects as distribution rather than as the place the value is stored.

There is no designed failure state. Camera permission is declined, the light is bad, or the phone is three years old, and the visitor gets a black rectangle on the product page — worse than never offering try-on at all. The degraded path is designed first and the clever path second, so the page still does its job for the customer whose device cannot run it.

See where the site stands before you brief anyone.

If the site is already live, start from what it scores rather than from a guess. The audit reports what it checked, what it found, and — the part most tools leave out — what it could not check, and why.

See the website audit Talk about this work

Questions people ask before they commit.

Does the customer have to install an app?: For a filter on Snapchat or TikTok they are already inside that app, which is the point of publishing there. For try-on on your own site, no install is required at all — it runs in the browser, asks for camera permission, and works from a link you can put in an email, an ad or a product page. Removing the install step is usually the largest single gain in how many people actually try it.
Do our customer photos get uploaded anywhere?: No. The camera frames are processed inside the visitor's browser and nothing is sent to a server, which also means no image storage, no retention question and a much shorter conversation with your legal team. We state that plainly on the page itself, because people hesitate before granting camera access, and what they hesitate over is where the pictures go — so the page answers that before it asks.
How accurate is the sizing on different faces?: The model is not placed at a fixed size. Position and rotation come from the line between the eyes and the scale from the bridge of the nose, so the frame resizes itself to each face rather than assuming an average one. It is convincing rather than optical-grade, and it is meant to answer does this suit me, not what is my exact pupillary distance.
How many products can go into a try-on?: As many as you like, but each one is real work — its own model, its own materials and its own scale calibration against the face landmarks. Product count is the biggest single driver of the price. Starting with a focused set of bestsellers keeps the first build small enough to judge on its own conversion figures, and what those show tells you which products are worth adding next.
What happens on a cheap phone or in bad light?: Tracking degrades before it fails — the overlay drifts or the frame rate drops — so there is always a designed fallback rather than a broken screen. Usually that is a still-image mode or the standard product gallery, offered automatically with a clear message. The page still has to sell to the visitor whose phone cannot run the clever version.
Should we build on one platform and stay there?: No, and this is a hard-won point. Meta shut its Spark AR platform down in 2025, and every effect and every workflow tied exclusively to it went with it. Effects are built so the underlying assets and the concept are portable, and anything that has to live long-term belongs on your own site rather than inside someone else's tooling.

Often taken together.

Start with the brief, not the invoice.

Send what you are trying to achieve. You get a written scope, a price and a date — or an honest “this is not the right service for that”.

Talk about AR try-on & filters All Immersive, 3D & Emerging Tech services