AI automation
Virtual try-on for ecommerce: how it works, what it costs and whether it cuts returns
Virtual try-on shows a shopper a product on themselves before they buy, using an uploaded photo, a live camera or body measurements. Published plans work out to about $0.02–$0.20 per try-on. The returns evidence we could read comes from fit tools, not photo try-on. VITON13 runs a measured pilot from $113.
What is virtual try-on, and how do photo, AR and size tools differ?
Virtual try-on shows a shopper a product on themselves, or on a model close to them, before checkout. Three technologies share the name and need different inputs, so a price or result for one says little about another.
| Kind | What the shopper does | What the store supplies | Weak spot |
|---|---|---|---|
| Generative photo try-on | Uploads a photo, or picks a model, and gets an image back | Ordinary product photos; Google asks for one garment on a front-facing model, mannequin or flat lay, at least 512 × 512 px | Looks plausible, but fit is not measured |
| Live AR try-on | Turns on the camera and sees eyewear, makeup, jewellery or headwear on face or hand | A 3D model per product (eyewear, jewellery, furniture): GlamAR prices its creation by category and SKU count, Banuba says digitisation is free | Needs the 3D asset first; neither vendor’s try-on list names clothing |
| Size and fit tools | Enters height, weight and gender, or measurements, and sees an avatar or a size tip | Size charts and, ideally, returns history | Shows fit, not how the item looks |
Google’s tool runs on Google, not on your page: its generative model combines photos of real models, shot by Google, with your garment photos. Merchants with a shopping feed and good images are opted in automatically.
What does virtual try-on cost per try-on?
Vendors sell allowances, not single try-ons, so the table divides each plan’s price by what it includes. The arithmetic is ours, at full use.
| Option | Price as published | Included | Per try-on at full use |
|---|---|---|---|
| Google Vertex AI Virtual Try-On API | $0.06 per image | One person image and one clothing image per call | $0.06 |
| Genlook widget (Shopify, WooCommerce, PrestaShop) | Free, then $19.99, $29 and $99 a month | 10 try-ons free; 100, 250 and 1,000 on paid plans; extras $0.17, $0.12 and $0.10 | About $0.20, $0.12 and $0.10 |
| Banuba Easy Virtual Try-On (AR, no-code) | $49, $99 and $349 a month; its Shopify extension is priced separately, at $319–$1,599 | 1,000, 3,500 and 15,000 try-ons | About $0.05, $0.03 and $0.02 on the Easy plans |
| GlamAR AR try-on (India) | Quotation only | Plans by monthly views (10,000, 50,000, 100,000) and SKU limit (50, 100, 500); assumes AR-ready 3D models | No price published |
| Google try-on in Search and Shopping | The help page lists no fee | Automatic for qualifying apparel images | Not applicable |
The tool fee is only part of the bill. At 1,000 try-ons a month the table gives about $60 (Google API), $99 (Genlook Pro) or $49 (Banuba Easy Basic); integration is a separate project. Three lines set the real bill:
- Volume: every generated image costs compute, so a sale that sends many shoppers to the button costs more than a quiet month.
- Assets: photo try-on reuses your product photos; AR needs a 3D model per product, and GlamAR says onboarding takes 2–4 weeks depending on whether those assets exist.
- Work around the widget: consent screens, analytics, mobile testing and a person checking results.
Vendors disagree on scope: Banuba lists free product digitisation, while GlamAR sells 3D model creation separately. Ask what the price includes.
Does virtual try-on really cut returns and lift sales?
The strongest evidence we could read is an academic randomized field experiment. Gallino and Moreno (Manufacturing & Service Operations Management, September 2018) randomly gave some online shoppers virtual fit information and compared them with a control group. Their abstract reports higher conversion rates and order values, and lower costs from returns and from ordering several sizes.
Two limits apply. The study predates generative photo try-on, and its abstract gives a direction, not a number you can copy into a budget.
Retailers publish bigger numbers with caveats. In October 2024 Zalando’s director of product size and fit said the company had seen up to a 40% reduction in return rates with its virtual fitting room, and added that the figure was limited by the testing phase. That tool builds an avatar from the shopper’s measurements, not from a photo.
Vendor blogs print larger figures, usually with no study or method behind them, so we repeat none. We found no randomized field study of generative photo try-on, which means its effect on your catalogue is unmeasured until you measure it.
What happens to the shopper’s photo, and which laws apply?
A try-on photo is personal data, and some regimes treat face or body data as biometric. This is not legal advice; have a lawyer review the consent text for each market.
| Place | What the source says | What it means for a widget |
|---|---|---|
| EU (GDPR) | Facial images count as biometric data when specific technical processing allows unique identification; processing them to identify someone is prohibited without an exception such as explicit consent (Article 9). Recital 51: a photograph is biometric only when processed that way. | Ask explicit consent before upload; use the photo only for the preview |
| Illinois (BIPA) | A February 2023 law-firm alert reports class actions over try-on tools at Estée Lauder, Louis Vuitton, Dior, Pandora and Wella, alleging no written notice, release or retention policy; courts split on dismissal. Damages: $1,000 per negligent and $5,000 per reckless violation. | Written notice, consent and a published retention rule before launch |
| India (DPDP) | The DPDP Rules 2025, notified 14 November 2025 with an 18-month phase-in, require standalone plain-language consent notices stating the purpose, and verifiable consent for a child’s data. | A separate notice, minimal data, verified consent for minors |
Vendors state their own retention: Genlook says customer photos are deleted after 7 days and never train models. Put such claims in the contract, with server locations.
- Offer a path with no upload, such as pre-generated images on models of different body types.
- Show the notice and an unticked consent box before the upload button works.
- Label every result as an AI-generated preview.
- Delete the photo once the preview exists unless the shopper keeps it.
Where does virtual try-on get it wrong?
- Fit: none of the vendor pages we read publishes an accuracy figure for fit. Treat a preview as colour, proportion and style, not a size recommendation.
- Coverage: Google enables try-on only for images that qualify: one garment, fully visible, on a front-facing model or flat, with no hands or bags covering details.
- Bodies: shoppers pick from models sized XXS to XXXL on Google, but a stand-in body is not theirs. Test with plus-size, petite and tall shoppers before trusting a demo.
- Category: AR suits what Banuba lists, such as eyewear, makeup and jewellery; photo try-on suits garments. A tool built for one rarely transfers to the other.
Should you use a plug-in, an API or the platform’s own try-on?
- Platform tool (Google): nothing to build and no fee on the help page, but it runs on Google’s page, only for apparel that qualifies.
- Plug-in widget (Genlook, Banuba and similar): live quickly, priced by volume, and the shopper stays on your product page. Check retention, branding and what happens past the plan limit.
- API (Google Vertex AI and vendor APIs): you design and maintain the experience, the consent and the data flow. It fits when you already have developers; custom 3D suits only a small, stable catalogue.
Pick by five checks: product category, store platform, how fast the catalogue changes, whether you have developers, and how much privacy work you can carry.
When should you skip virtual try-on?
- The product is bought on specifications, not on looks or fit.
- Your returns come from damage, quality or late delivery rather than size or appearance.
- You have too few orders to read a before-and-after difference.
- You sell on a marketplace, which does not accept custom widgets on product cards.
- Your product photos are cluttered or inconsistent and you would not reshoot them.
- You cannot take on the consent and privacy work.
What would a try-on pilot cost with VITON13?
A pilot is priced as a project, separate from any vendor or interface fee.
Cost calculator
Plan your try-on pilot
Tick what your store needs; the totals update as you choose.
Your estimate
- AI-assisted
- from $113
- Human-led
- from $173
- Timeline
- 10–18 working days
stages can overlap; the written estimate fixes the dates
Get a written estimate in 1 working dayStarting prices of VITON13 packages; the written scope fixes products and the final price after the brief. Vendor fees are paid separately.
How VITON13 does it
This is a measured pilot, not a catalogue rollout: up to 10 products on a store whose product pages can be edited (Shopify, Next.js, WooCommerce or similar). We check your category and platform, compare three methods (a third-party app or SDK, pre-generated AI model images with no shopper photo, or photo upload through a vendor’s interface), connect the fitting one with a consent step and an AI label, and add analytics. In the AI-assisted mode AI tools produce setup and renders and a developer checks them before launch; in the human-led mode a stylist or product specialist also reviews every render and writes size guidance for the 10 products.
Virtual try-on for online stores
- Price
- $113 AI-assisted or $173 human-led
- Timeline
- 10–18 working days
The final price is fixed in writing after a short scope review.
AI look preview
- Price
- $33 AI-assisted or $53 human-led
- Timeline
- 2–3 working days
Ecommerce development
- Price
- $113 AI-assisted or $173 human-led
- Timeline
- 10–18 working days
AI-assisted: AI drafts inside defined steps, a person directs and checks every result. Human-led: a specialist does the work and AI assists.
What is included
- A method and platform check with one recommended method, its costs and limits
- Try-on on up to 10 product pages, with a consent step and an AI-generated preview label
- At most 4 renders per product on 2–3 body types and skin tones if the no-shopper-photo method is chosen
- Four analytics events, a two-week measurement plan and an off switch
- Delivery: 10–18 working days after the brief is approved
Not included
- Vendor or interface fees, paid by you directly
- 3D modelling of products, which is quoted separately
- More than 10 products, and marketplace product cards
- Legal approval of your privacy policy, which your counsel gives
- Any promise about conversion or return rates
Revisions: 2 revision rounds on the entry button and the try-on window
Get an estimate in 1 working dayWritten estimate within 1 working day
Example
PILOTPilot in progress
No published try-on pilot yet
VITON13 has not run a try-on pilot for a client, and no try-on vendor has been chosen. A test on up to 5 men’s products in our own Old Money store, store.viton13.com, is planned and has not started, so this guide shows no results.
What this does not prove
- The price table is published list prices plus our arithmetic. It does not show what a pilot would do to your returns or sales.
Questions and answers
Does Google charge merchants for its apparel try-on?
Google’s Merchant Center help page lists no fee. Brands with a shopping feed and good images are opted in automatically, and paid ads can opt out of try-on assets through Google Customer Support. It runs on Google’s surfaces, so it does not replace a try-on on your own product page.
How long does it take to add virtual try-on to a store?
Vendors quote anything from a few minutes for a no-code Shopify app (Genlook) to two to four weeks for AR onboarding (GlamAR). VITON13’s pilot takes 10–18 working days after the brief is approved, then two weeks of measurement.
What should the consent screen say?
In plain words: what the photo is used for, how long it is kept, who processes it and how to delete it, in a notice separate from the terms. India’s DPDP Rules ask for a standalone notice stating the purpose; GDPR needs explicit consent for special categories. Not legal advice.
Which numbers should I track in a pilot?
Try-on opened and completed per product, add to cart after try-on, purchases, and returns on the same products if your platform records them. Compare randomly assigned groups where you can. Four events over two weeks give a first reading, not proof.
Can I use virtual try-on on Amazon or other marketplace listings?
Not as a widget. Marketplaces do not let sellers add custom widgets to product cards, and Google’s tool runs on Google’s own pages. AI model pictures for a card are a different job.
Sources
- Google Merchant Center Help, About apparel virtual try-on —
- Google Cloud, Vertex AI pricing (Virtual Try-On) —
- Genlook, Pricing and FAQ —
- Banuba, Virtual try-on pricing plans —
- GlamAR, AR try-on pricing —
- Gallino and Moreno, The Value of Fit Information in Online Retail (M&SOM 20(4)) —
- Zalando, enhanced virtual fitting room with 3D avatar —
- EUR-Lex, Regulation (EU) 2016/679 (GDPR) —
- Arent Fox Schiff, Biometric privacy class actions take aim at virtual try-on retailers —
- PIB India, Government notifies DPDP Rules —