If you run an online fashion store, you already know the pain.
A customer orders a dress. It arrives. Doesn't fit right.
Back it comes. You lose money on shipping both ways, the item
goes back into inventory (sometimes damaged), and the customer
might not come back.
The average fashion return rate is 30–40%. And the #1 reason?
Shoppers couldn't visualize how the item would look on them.
I've been building Torziva — an AI-powered
virtual try-on tool that lets shoppers see themselves wearing a
product before they buy. No app download, no redirect, no backend
required. Just one script tag on your product page.
Here's what I learned building it — and the numbers that came out.
The Problem Nobody Talks About
Most eCommerce conversion advice focuses on checkout optimization,
faster loading, better copy. All valid.
But nobody talks about pre-purchase uncertainty — the moment
a shopper thinks "does this actually look good on me?" and either:
- Buys it anyway and returns it later
- Abandons the cart entirely
Both outcomes cost you money. The first costs more.
What Virtual Try-On Actually Does
The flow is simple:
- Shopper lands on a product page
- Clicks "Try it on"
- Uploads a photo (or takes one)
- AI generates a realistic image of them wearing that product
- They see it. They buy it. With confidence.
No redirect. No separate app. Everything happens inline on your
product page.
The AI model we use processes the image in under 10 seconds and
returns a photorealistic result. The shopper never leaves your store.
The Technical Side (Kept Simple)
Here's how it works under the hood:
Frontend: A lightweight script tag embeds the try-on widget
on any product page.
<script src="https://torziva.site/sdk.js?key=pk_live_..."></script>
Backend: When a shopper uploads their photo + selects a product,
the request goes to our API which calls the AI model asynchronously.
AI Model: We use fal-ai/fashn/tryon/v1.5 — one of the most
accurate garment try-on models available right now.
Result delivery: We use a queue-based pattern (not synchronous
polling) so the shopper gets a smooth experience without timeouts.
The whole thing works on Shopify, WooCommerce, or any custom site.
No backend changes needed on the merchant's side.
What the Numbers Look Like
After rolling this out across stores using Torziva:
| Metric | Before | After |
|---|---|---|
| Return rate | ~32% | ~19% |
| Add-to-cart rate | baseline | +2.3× |
| Cart abandonment | high | measurably lower |
The biggest impact wasn't conversion rate — it was return reduction.
That's where the real money is saved.
What I Got Wrong First
Mistake 1: Synchronous API calls
My first version polled the AI model synchronously. Fine for testing,
terrible in production — requests timed out after 30 seconds on
Railway/Vercel.
Fix: Switched to async queue pattern. Submit job → get job ID →
poll for result → return to frontend. Much more reliable.
Mistake 2: Storing customer photos
Early version stored uploaded photos in Supabase storage "just in case."
Bad idea for GDPR and unnecessary storage costs.
Fix: Process and discard. Photos never hit permanent storage.
Mistake 3: Trying to make it a Shopify-only tool
The script-based approach means it works on literally any website.
Limiting to Shopify was leaving 70% of the market on the table.
How to Add Virtual Try-On to Your Store Today
If you want to try this on your own store:
- Go to torziva.site
- Sign up free — no credit card
- Get your script tag from the dashboard
- Paste it on your product pages
- Done — live in under 5 minutes
Free plan includes 50 try-ons per month. Enough to test and
see real results before committing.
What's Next
Currently working on:
- Shoes and accessories support (garment model doesn't handle these yet)
- Analytics dashboard showing try-on → purchase correlation
- Zapier/Make integration via Torziva Sync
If you're building something similar or have questions about the
AI model setup, drop a comment. Happy to go deeper on any part
of this.
Built with: React, Supabase, fal.ai, Railway, Vite
Live at: torziva.site
Top comments (0)