~/work/stylesyncs
StyleSyncs
AI-powered virtual try-on for fashion
A generative-AI fashion visualization platform that lets users virtually try on clothes. Grew to 100+ registered users and raised $9K+ through competitive pitch competitions.
React
TypeScript
Supabase
Gemini API
GCP Cloud Run
Python100+
registered users
$9K+
raised in pitch competitions
4+
specialized AI models in the pipeline
stylesyncs.com
01 · the problem
Online shoppers can't tell how clothes will actually look on them, and that uncertainty drives returns, abandoned carts, and frustration. StyleSyncs set out to close that gap with generative AI: upload a photo, pick a garment, and see yourself wearing it.
02 · what I built
- Co-founded the company and lead all AI development, from model selection and pipeline architecture to production deployment.
- Architected a multi-model AI pipeline across 4+ specialized models, each handling the part of the problem it's best at.
- Built the product on React + TypeScript with Supabase for auth and data, deployed on GCP Cloud Run.
- Currently developing a real-time video try-on feature for interactive garment overlay on live video feeds.
03 · the pipeline
- 1
ingest
User uploads a photo and selects a garment through the React/TypeScript frontend, with Supabase handling auth and storage.
- 2
analyze
A multimodal Gemini model analyzes the clothing item (category, fit, texture) to condition the downstream generation.
- 3
avatar
Gemini image models create a clean digital avatar from the user's photo.
- 4
generate
IDM-VTON, a diffusion-based virtual try-on model, renders the garment onto the avatar with realistic drape and lighting.
- 5
serve
Results are served from GCP Cloud Run back to the user's gallery.
04 · key decisions
Specialized models over one general model
No single model handles clothing analysis, avatar generation, and try-on rendering well. Splitting the pipeline across 4+ specialized models (diffusion for try-on, multimodal LLMs for analysis) gave us markedly better output quality at each stage.
Ship to real users early
We launched before the pipeline was perfect and let real usage drive priorities. That's how real-time video try-on became the next bet; users kept asking for it.
05 · where it stands
- ✓100+ registered users on the live platform at stylesyncs.com.
- ✓$9,000+ raised through competitive pitch competitions.
- ✓Real-time video try-on in active development.
ReactTypeScriptSupabaseGemini APIGCP Cloud RunPython