README.md

metadata

title: Ai_Clothes_change
sdk: docker
emoji: ⚡
colorFrom: yellow
colorTo: blue

✦ AI Clothes Changer — Virtual Try-On

AI-powered virtual try-on app using CatVTON diffusion model. Upload a person photo + a clothing image → get a photorealistic try-on result in seconds.

---

🧠 Tech Stack

Layer	Technology
AI Model	CatVTON (diffusion-based VTON)
Backend	Python · FastAPI · PyTorch
Frontend	React · Vite · Tailwind CSS
Inference	CUDA (GPU) / CPU fallback
Packaging	Docker · docker-compose

---

📁 Project Structure

ai-clothes-changer/
├── backend/
│   ├── app/
│   │   ├── main.py              # FastAPI app + CORS
│   │   ├── routes/tryon.py      # POST /api/v1/try-on
│   │   ├── services/
│   │   │   ├── inference.py     # CatVTON model loading + inference
│   │   │   ├── preprocessing.py # Image resize, normalize
│   │   │   └── postprocessing.py# Tensor → PIL → base64
│   │   ├── schemas/request.py   # Pydantic models
│   │   └── utils/
│   │       ├── config.py        # Settings from .env
│   │       └── file_handler.py  # Upload + cleanup helpers
│   ├── weights/catvton/         # ← model weights go here
│   ├── requirements.txt
│   └── .env
│
├── frontend/
│   └── src/
│       ├── api/api.js           # Axios client
│       ├── components/
│       │   ├── Upload.jsx       # Drag-and-drop uploader
│       │   └── Result.jsx       # Output + download
│       └── pages/Home.jsx       # Main UI
│
├── scripts/
│   ├── download_model.py        # Downloads CatVTON weights
│   └── train.py                 # Fine-tuning stub
│
├── docker/Dockerfile
└── docker-compose.yml

---

🚀 Quick Start

1. Download model weights

python scripts/download_model.py

This downloads ~4GB of weights from Hugging Face into backend/weights/catvton/.

2. Backend (Python)

cd backend
python -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8000

API docs: http://localhost:8000/docs

3. Frontend (React)

cd frontend
npm install
npm run dev

Open: http://localhost:5173

---

🐳 Docker (Production)

# Build and run backend
docker compose up --build

# Dev mode (backend + frontend)
docker compose --profile dev up --build

---

🔌 API Reference

POST /api/v1/try-on

Field	Type	Description
person_image	File	Full-body person photo (JPEG/PNG/WEBP, max 10MB)
cloth_image	File	Clothing item photo
cloth_type	string	upper | lower | overall
num_steps	int	Inference steps 20–100 (default: 50)

Response:

{
  "status": "success",
  "output_image": "/outputs/results/abc123.png",
  "output_image_base64": "data:image/png;base64,...",
  "processing_time_ms": 4200.5,
  "message": "Try-on completed successfully"
}

GET /api/v1/health

{ "status": "healthy", "model_loaded": true, "device": "cuda" }

---

⚙️ Configuration (.env)

DEVICE=cuda            # cuda | cpu
MODEL_PATH=weights/catvton
MIXED_PRECISION=fp16   # fp16 | no
OUTPUT_IMAGE_WIDTH=768
OUTPUT_IMAGE_HEIGHT=1024
CLOTH_TYPE=upper
MAX_IMAGE_SIZE_MB=10

---

💡 Tips for Best Results

• Person image: Full-body photo, neutral background, front-facing
• Cloth image: Flat-lay or product photo on white background
• Steps: 50 = balanced quality/speed · 80+ = best quality (slower)
• GPU: CUDA speeds up inference 10–20× vs CPU

---

🗺️ Roadmap

• Cloth library / preset gallery
• Video try-on (CatV2TON)
• Background removal (rembg)
• Body size / pose adjustment
• Cloudinary / S3 storage
• Redis queue for concurrent requests
• Mobile-optimized UI

---

📄 License

MIT — built on CatVTON (Apache 2.0).