Fitting Room

Uploading your photo to FittingRoom
Searching for a preppy style using AI
Custom Outfit Uploader
An example of our AI Outfit Creator with "preppy" style
A example outfit we found
Example of trying on the preppy outfit

Inspiration

We wanted to shrink the gap between discovering a style and actually seeing it on you. Shopping sites show product photos, but it’s hard to imagine how items fit you. We built FittingRoom so people can virtually test real items from the web, driven by AI style understanding.

What it does

Upload a user photo.
Enter/choose a style and get AI-generated shopping queries.
Scrape top product results via SerpApi + Amazon and enrich them with metadata.
Generate realistic virtual try-on composites using Vertex AI (real product images overlaid on user photo).
Present side-by-side comparison and direct “Buy This Outfit” links.

How we built it

Frontend: Next.js + React + TypeScript
Styling: Tailwind CSS
Backend: Next.js API Routes (serverless)
Style analysis: OpenAI ChatGPT (image-to-text prompts + query generation)
Product search: SerpApi (Amazon product scraping) and parsing results to extract images, categories, colors, sizes, and links
Virtual try-on / image composition: Google Vertex AI for realistic composition of user photo + product images

Challenges we ran into

GCP configuration complexity: setting up Vertex AI, service accounts, permissions, etc. took longer than expected.
API changes: Google Shopping’s individual product pages were deprecated shortly before the hack, so we pivoted to SerpApi + Amazon to retrieve product images and shopping links.
Compositing realism: producing realistic overlays that preserved original person while also using the new outfits.

Accomplishments that we're proud of

Full end-to-end flow: upload → AI style analysis → product discovery → virtual try-on → purchase links.
Real-product composites: our Vertex AI compositions maintain skin tone and lighting well enough that outfits look believable in side-by-side comparisons.
Smart query generation: ChatGPT-generated queries produced relevant product matches across diverse styles (punk → studs, vintage → high-waisted denim, minimal → neutral staples).

What we learned

The fragility of third-party product pages since product sources change often and you must plan for fallback data providers.
Prompt engineering is critical: small prompt adjustments greatly affect how Vertex AI composites product images onto photos.