Custom Diffusion Web App

A clean, standalone Gradio web interface for high-fidelity Image-to-Image transformations using a custom-trained Diffusion Model.

Project Structure

├── app.py                   # Main Gradio web interface
├── model_server.py          # PyTorch inference wrappers, UNet model, and SDEdit scheduler
├── final_enhanced.ipynb     # Complete training and data augmentation pipeline
├── requirements.txt         # Python dependencies
└── models/                  # Directory for pre-trained weights (ignored by Git)

Setup Instructions

1. Create a Virtual Environment

Create and activate a Python virtual environment to manage dependencies cleanly:

Windows (PowerShell):

python -m venv .venv_gradio
.\.venv_gradio\Scripts\activate

Linux / macOS:

python3 -m venv .venv_gradio
source .venv_gradio/bin/activate

2. Install Dependencies

Install the required PyTorch, torchvision, and Gradio packages:

pip install -r requirements.txt

3. Provide Model Weights

Due to GitHub file size constraints (>100MB), the model weight checkpoints are not included in the repository repository history.

Ensure you place your pre-trained model checkpoint inside the models/ directory:

• Expected path by default: models/diffusion_model_street.pth

Note: If you are starting from scratch, run the training pipeline inside final_enhanced.ipynb to train your model and generate the checkpoint files.

Running the Web Application

Launch the Gradio application locally:

python app.py

Or using the direct path to the virtual environment's Python executable:

.\.venv_gradio\Scripts\python app.py

Once loaded, the terminal will display a local web server URL (typically http://127.0.0.0:7860). Open this link in your browser to access the interactive diffusion editing suite.

Features

• SDEdit Image-to-Image Transformation: Upload an image and adjust the Strength slider to control how much noise is injected before the custom UNet denoises the structure back into your model's target domain.