Synthetic-Image-Detector

Synthetic-Image-Detection A deep learning model built with TensorFlow/Keras and deployed as a Streamlit app to classify images as either real photographs or AI-generated.

This model (best_real_vs_ai_detector.h5) detects whether an image is AI-generated or Real.
It is a fine-tuned VGG16 model trained on custom datasets of AI-generated and real images.

---

🔍 Overview

The model outputs a single probability score:

• Values > 0.5 → REAL
• Values ≤ 0.5 → AI-GENERATED

It is used in an interactive Streamlit web app that allows users to upload an image, get predictions, and provide feedback to continuously improve the system.

---

🧩 Model Architecture

• Base: VGG16 (pretrained on ImageNet)
• Added Layers:
  • Flatten
  • Dense(512, ReLU)
  • Dropout(0.5)
  • Dense(1, Sigmoid)
• Loss Function: Binary Crossentropy
• Optimizer: Adam (learning rate = 1e-4)
• Input Size: 150 × 150 × 3 RGB
• Output: 1 neuron (binary classification)

---

📂 Folder Structure

project/
│
├── app.py                  # Streamlit web interface
│
├── models/
│   └── best_real_vs_ai_detector.h5   # Model file (auto-downloaded from Hugging Face)
│
├── feedback/               # Saved feedback images from the Streamlit app
│   ├── correct_real/
│   ├── correct_ai/
│   ├── incorrect_real/
│   └── incorrect_ai/
│
├── requirements.txt        # Dependencies for app and scripts
└── README.md               # Documentation

💻 Usage
1️⃣ Install dependencies
pip install -r requirements.txt

2️⃣ Run the Streamlit app
streamlit run app.py


The first run will automatically download the model from Hugging Face into the models/ folder.

Then open the app in your browser:
http://localhost:8501


Upload an image, click Analyze, and the app will tell you if it’s Real or AI-Generated — with confidence percentage.

You can also provide feedback (Yes/No) to help future retraining.