🧠 Real vs AI-Generated Image Detection System

This project aims to detect whether a given image is real (photographed) or AI-generated using deep learning techniques. It is built from scratch for better interpretability and understanding of how such classification models work.

🚀 Project Overview

With the rise of generative models like DALL·E, MidJourney, and Stable Diffusion, distinguishing between real and AI-generated images has become a major challenge. This system tackles that problem using a custom-built Convolutional Neural Network (CNN).

✅ Features

• 🛠 Built from scratch (no pretrained models used)
• 📊 Custom CNN architecture for explainability
• 📁 Dataset split into Real and AI-Generated images
• 📈 Visualization of training progress (accuracy/loss)
• 🧪 Evaluation using accuracy, precision, recall, F1-score
• 📷 Accepts image input and predicts its authenticity

🧰 Tech Stack

• Python
• NumPy, Pandas
• Matplotlib, Seaborn
• TensorFlow / Keras
• Scikit-learn

📂 Project Structure

Real-vs-AI-Image-Detection/
│
├── dataset/
│   ├── real/              # Real images
│   └── ai/                # AI-generated images
│
├── model/
│   └── cnn_model.py       # Custom CNN architecture
│
├── notebook/
│   └── training.ipynb     # Jupyter notebook for training & evaluation
│
├── utils/
│   └── preprocessing.py   # Image preprocessing functions
│
├── saved_model/
│   └── model.h5           # Trained model weights
│
├── predict.py             # Prediction script for a single image
├── requirements.txt       # List of required packages
└── README.md              # You're reading it now!

🏁 Getting Started

1. Clone the repository

git clone https://github.com/OshankAgrawal/Real-vs-AI-Generated-Image-Detection-System.git
cd Real-vs-AI-Image-Detection

2. Install dependencies

pip install -r requirements.txt

3. Train the model

Run the training notebook or cnn_model.py script to train the model using the provided dataset.

4. Predict on a new image

python predict.py --image path_to_your_image.jpg

📊 Results

Metric	Value
Accuracy	64.93%
Precision	65%
Recall	65%
F1-Score	65%

🙋‍♂️ Author

Oshank Agrawal B.Tech – Artificial Intelligence & Data Science Samrat Ashok Technological Institute, Vidisha (M.P.) 🌐 LinkedIn • 📧 oshankagrawal@gmail.com

📌 Note

This project is for educational and research purposes. Model performance may vary depending on dataset quality and size.

⭐ Star This Repo

If you found this project helpful or interesting, consider giving it a ⭐ on GitHub to support the work!