Is virtual try-on free? What costs nothing in 2026 — and what you pay instead

"Free virtual try-on" is one of the most typed phrases in this whole category, and the honest answer has three parts: some try-on is genuinely free forever, some is free until a counter runs out, and some is free in the way a printer is cheap. Here is which is which — and the arithmetic that explains every paywall in the category.

Published 4 September 2026 · Written by the team building Hoilo

The short answer

Three kinds of AI try-on cost nothing and will keep costing nothing:

• Try-on built into a shop — the retailer pays for it because it sells the garment.
• Try-on inside Google Search and Shopping — same logic, one layer up.
• Open-source models run as public demos — somebody else's GPU, offered as a shop window for the research.

Everything else — every standalone app and every "try on clothes online free" website — meters you. It might be three renders a day, twenty credits on signup, one free result before a paywall, or a free tier that quietly halves your resolution. The meter is not greed; it is arithmetic, and once you have seen the numbers the whole market makes more sense.

1. Free because it sells clothes

The best-quality free try-on in 2026 is the kind a retailer bolted onto its own catalogue. ASOS launched try-on inside its iOS app with the AI platform AIUTA in February 2026 — roughly ten thousand products at the start, a result in a few seconds, and a choice between uploading your own photo or picking an AI-generated model built to resemble you. It costs nothing, and it never will, because the entire point is to reduce the number of jackets that come back.

This is also the best render you will get anywhere. The retailer owns lit studio photography of the garment at high resolution, which is exactly the input an image model wants. A phone photo of a rail in a shop cannot compete with it.

The limit is coverage. It works on that retailer's enabled products, in that retailer's app, in the regions where it launched. Comparing a coat there with one somewhere else means two apps and two mental notes.

2. Free because Google pays for it

Google folded try-on into Search and Shopping listings — and shut the standalone Doppl app down on 30 April 2026, less than a year after launching it. The message was clear enough: try-on is a feature of shopping, not a destination.

For a shopper that is good news. You upload a photo of yourself once, then tap try-on straight from product results across an enormous slice of the shopping index, using the same generation model that powers Google's other image features. No install, no subscription, no credits, and no account beyond the Google one you already have.

What you give up is reach and control. If the garment has no listing in Google's index — a small independent label, a piece from Vinted, a screenshot a friend sent you, the dress on the rail in front of you right now — there is nothing to tap. And the feature reaches countries on Google's schedule, not yours.

3. Free because a researcher published the model

The genuinely-free-and-nobody-is-selling-you-anything option is the open-source one. Several strong garment-transfer models are published openly, and their authors usually put up a public demo you can use in a browser without an account — the Hugging Face Spaces running the well-known try-on models are the usual example.

You get: no signup, no card, no credit counter, no watermark.

You pay in patience and polish. Public demos share one small pool of GPUs, so you queue behind everyone else who found the same link, and popular Spaces go down or get rate-limited for days. The workflow is a research demo — upload a person image, upload a flat garment image, wait — not a phone app, so it wants a clean product cutout rather than a screenshot with a model, a price and a Buy button in it. Results are typically fine on plain garments and weak on prints, layers and anything sheer.

Use it when: you want one or two try-ons, you have a clean garment image, and you would rather queue than pay or sign up for anything.

Why nobody gives away unlimited try-ons

Here is the number that explains the whole market. Almost every app in this category calls a hosted image model rather than running its own GPUs, and at 2026 list prices those calls cost roughly four to fifteen cents per generated image, depending on the model and the output resolution. Google's widely-used image models — the family people call "nano banana" — sit right in that band, with the 2026 flash generation around six or seven cents at default resolution and the pro model several times that.

One try-on	A few cents of model time, plus storage, bandwidth and the retries when a first render comes out wrong.
One curious user	Rarely stops at one image. Ten to thirty renders in a first session is normal — call it a dollar or two.
Ten thousand of them	Five figures a month in model spend alone, before a single person has paid anything.

So an app has exactly four ways to survive: charge you, cap you, sell something else about you, or die. Most of the "unlimited free AI try-on" sites that appear at the top of search results are picking option four without knowing it, which is why so many of the tools in last year's listicles now return a parked domain.

This is also why a free tier is a reasonable thing to see, rather than a trick. A handful of free renders costs a developer real money and lets you find out whether the thing works on your photos before paying. Be more suspicious of unlimited than of limited.

Five things "free" costs instead of money

1. A watermark. The honest version. You can see it, and you knew about it before you uploaded anything.
2. Your email, then your inbox. Free-credits-on-signup is a lead magnet. Assume the address is worth more to the site than the credits it cost.
3. Your photo, kept. The one that matters. A free tool still has to justify its GPU bill, and retained user photos have value — as training data, or as a dataset someone sells later. This is the cost you cannot see and cannot undo.
4. Quality, quietly. Free tiers often route to a cheaper model or a smaller output, then upsell you on "HD". You are not comparing the tool's quality; you are comparing its cheapest setting.
5. The render-then-paywall trick. The result generates, blurs, and asks for a card to reveal it. It works because you already invested a photo and a wait. Treat it as a sign of how the rest of the product will treat you.

What to check before you upload a photo to anything free

1. Find the retention sentence. Not the privacy page — the actual sentence about how long images are kept. If nobody wrote one, assume forever.
2. Find the training sentence. "We may use uploaded content to improve our services" means your photo is training data. That may be fine with you; it should not be a surprise.
3. Check what happens at zero credits. Does the app tell you the price up front, or after it has your photo and your result?
4. Test the hard garment first. Free renders are limited, so do not spend them on a plain t-shirt. A bold print, a logo or a sheer layer tells you in one image whether the tool is any good.
5. Run the same photo twice. Generation is stochastic. If two runs disagree wildly, no single result is worth trusting — and you have learned that for free.

Our own guide on trying clothes on from a photo covers which photo of yourself to use, which will waste a credit, and why.

What we would actually do

• The garment is at a big retailer with try-on. Use theirs. Free, and the best render available.
• The garment is mainstream and indexed. Google's try-on in Search or Shopping. Free, no install.
• You need one try-on, ever, and have a clean product image. An open-source demo. Free, queue and all.
• The garment is a screenshot, a small label, or something on a rail in front of you. None of the free options can reach it. This is the case a standalone app exists for — and the case you should expect to pay something for, because someone is paying for every image either way.
• You will not upload a photo of yourself anywhere. Look for the option ASOS shipped: an AI-generated model built to resemble you, rather than your own face. Then read where your photo goes.

Questions people ask