💃

AI Dress Try-On App — Avnish Yadav

A virtual fitting room powered by Google Gemini Nano and Banana AI — upload your photo and any dress to see a realistic try-on result instantly.

← Back to Projects

Overview

The AI Dress Try-On App is a virtual fitting application that removes the guesswork from online clothes shopping. Users upload a photo of themselves and an image of any dress, and the system generates a realistic composite showing how the outfit would look on them — complete with smart styling suggestions and personalized recommendations.

"Online shopping's biggest problem is fit uncertainty. This app brings the fitting room to your browser using the latest generative AI models."

Features

  • Upload a personal photo and a dress image — get a realistic try-on result
  • Powered by Google Gemini Nano for on-device AI inference
  • Banana AI for scalable cloud-based image generation
  • Smart styling suggestions based on the outfit and user profile
  • Personalized recommendations for complementary items
  • Clean, intuitive UI — no technical knowledge required

Tech Stack

Google Gemini Nano Banana AI JavaScript Generative AI Image Processing REST APIs

The app combines on-device AI (Gemini Nano) for fast, private inference with Banana AI's cloud infrastructure for high-quality image generation at scale. This hybrid approach balances speed, quality, and cost.

How It Works

Input Processing

The user uploads two images: a photo of themselves and an image of the desired dress. The app preprocesses both images — normalizing dimensions, extracting body pose data, and identifying garment structure.

AI Generation

Gemini Nano handles the understanding layer — analyzing body shape, lighting, and garment fit. Banana AI's diffusion model generates the final composite image with realistic fabric draping and lighting.

Styling Suggestions

Based on the outfit analysis, the system generates complementary styling recommendations — accessories, shoes, and color pairings that work with the selected dress.

Personalization

The recommendation engine learns from user preferences and browsing patterns to surface more relevant suggestions over time.

What I Learned

  • Integrating on-device AI models (Gemini Nano) with cloud-based generation pipelines
  • Working with image-to-image diffusion models via API
  • Handling large binary payloads (images) efficiently in the browser
  • Designing UX for AI-powered features where output quality varies
  • Balancing latency vs. quality in generative AI applications

More details coming soon — this project is actively being developed.