Geni —The Autonomous Virtual Try-On Agent

Geni is an AI virtual stylist powered by Gemini & Tiger Data that swaps full head-to-toe outfits in one shot, pulls live retail prices, and lets anyone remix looks in a 1-click community closet.

Comment

Inspiration

Existing virtual try-on tools only swap a single shirt at a time, miss small items like shoes or accessories, and never tell you what the outfit actually costs. We wanted an autonomous stylist that transfers an entire head-to-toe look in one shot and tracks real-world prices.

What it does

Geni is an intent-aware AI virtual stylist and social closet. Users upload a model photo—or speak a budget and style request—and Geni detects every garment from head to toe, pulls live retail prices via Google Search, and dresses the user in the complete outfit at once. Users can also publish looks to the Geni Community Feed, where anyone can click "Try This On Me" to test the exact same fit on their own photo.

How we built it

  • Google Gemini API (@google/genai): Used gemini-3.8-flash with 2D spatial bounding boxes ([ymin, xmin, ymax, xmax]), structured JSON schemas, and Google Search Grounding to extract garments and live retail prices, paired with gemini-nano-banana-2.1 for full-body identity-preserving virtual try-on.
  • Tiger Data (PostgreSQL + Hypertables + pgvector): Stored shared community outfits, garment embeddings for similarity search, and real-time price/latency telemetry inside a Tiger Data hypertable.
  • Frontend & Tooling: Built with Next.js, TypeScript, and Tailwind CSS using Claude Code.

Challenges we ran into

Preserving the user's exact facial identity and body proportions while simultaneously swapping tops, bottoms, and footwear required careful multi-image conditioning and spatial bounding-box prompts rather than basic image masking.

Accomplishments that we're proud of

  • Achieving single-pass, head-to-toe outfit replacement (including shoes and small accessories) with an instant itemized retail receipt.
  • Building a sub-second "1-Click Try It On Me" community closet backed by Tiger Data hypertables and vector search.

What we learned

We learned how to chain Gemini's native multimodal vision, spatial coordinate grounding, and live Google Search tool calls into a deterministic agent router, alongside managing time-series and vector data in Tiger Data.

What's next for Geni —The Autonomous Virtual Try-On Agent

Adding WhatsApp/Telegram bot integration, real-time webcam mirror mode via the Gemini Live API, and 1-click instant checkout across retail partners.

Built With

Share this project:

Updates

Submission history