The problem I was trying to solve
Fashion ecommerce has a dirty secret — 30-40% of all purchases get
returned. The #1 reason? "It didn't look right on me."
Shoppers can't tell how a dress will fit their body from a flat
product photo. So they buy, try, and return.
I built Torziva to fix this.
What Torziva does
Torziva adds a "Try On" button to any fashion store's product page.
When a shopper clicks it:
- They upload their photo
- AI places the exact garment on their body
- They see a realistic on-model photo — in under 30 seconds
Real fabric texture. Real fit. Real colors. On their actual physique.
The tech stack
Here's what I used to build it:
- fal.ai — for the AI image generation / virtual try-on model
- Supabase — auth, database, storage
- Next.js 14 — frontend + API routes
- Vercel — deployment
The core flow is simple:
User uploads photo →
API sends to fal.ai with garment image →
fal.ai returns generated image →
Display result to user
Why fal.ai?
I evaluated a few image generation APIs. fal.ai won because:
- Fast inference (~30 seconds for try-on)
- Pay per generation ($0.075/image)
- Good try-on model quality for fashion garments
- Simple API, easy to integrate
The Supabase setup
I used Supabase for three things:
- Auth — email + Google OAuth for store owners
- Database — storing generation history, store configs
- Storage — saving generated images
Supabase MCP made the setup surprisingly fast.
Connected directly from Cursor and scaffolded the schema in minutes.
Challenges I ran into
1. Garment detection quality
The AI works best with flat lay photos on white backgrounds.
Dark backgrounds or heavy shadows confuse the model.
Had to add input validation and guidance for store owners.
2. Shopify integration
Getting the Try On button to appear correctly on Shopify product
pages required careful theme injection. Shopify's script tag API
made this manageable but took longer than expected.
3. Pricing
At $0.075/generation, pricing the product correctly was tricky.
Had to account for Shopify's 15% fee on top.
Ended up with a usage-based model that scales with store size.
Results so far
- Launched on Product Hunt on March 12
- Listed on Peerlist, BetaList
- First users onboarding now
Still early days but the feedback from fashion store owners has
been positive — especially the "no photoshoot needed" angle.
What's next
- Full Shopify app (in review)
- Support for more garment types
- Bulk generation for store catalogs
If you're building anything in fashion tech or ecommerce AI,
would love to connect. And if you run a fashion store —
try it free at torziva.site.
Happy to answer any questions about the tech stack in the comments 👇
Top comments (0)