README.md

COIN-TOSS: AI & Identity Risk Detection

Overview

COIN-TOSS is an advanced web application designed to accurately detect AI-generated images and assess potential identity theft risks. By combining multiple state-of-the-art deep learning models with custom analysis logic ("Gap Trap V3"), it provides a reliable "Real" vs "AI" verdict without ambiguous percentages, while also identifying potential misuse of authentic images.

Features

• High-Accuracy AI Detection:
  • Utilizes a hybrid ensemble of models (dima806/ai_vs_real_image_detection and prithivMLmods/Deep-Fake-Detector-v2-Model).
  • Gap Trap V3 Logic: A specialized algorithm to catch "uncanny valley" images and properly classify filtered real photos vs. high-quality deepfakes.
  • Frequency Analysis: Visualizes invisible noise patterns (FFT) to detect checkerboard artifacts common in diffusion models.
• Identity Theft Risk Analysis:
  • Analyzes "Real" images for biometric metrics (Face Visibility, Quality, etc.).
  • Provides a risk assessment (Low/High) for using the image in sensitive contexts (KYC, Profiles).
• User-Friendly Interface:
  • Simple drag-and-drop upload.
  • Instant "Real" or "AI" verdict.
  • Detailed analysis points explaining the decision.

Workflow

Prerequisites

• Python 3.8+
• Git

Installation

1. Clone the Repository

   git clone https://github.com/madhavmullick2025/COIN-TOSS.git
   cd COIN-TOSS
   

2. Install Dependencies It is recommended to use a virtual environment.

   pip install -r requirements.txt
   

Usage

1. Start the Application

   python app.py
   

   Note: The first run may take a few moments to download the necessary model weights from HuggingFace.

2. Access the Interface Open your web browser and navigate to:

   http://localhost:5002
   

3. Analyze Images

   • Upload an image (JPG, PNG, WEBP).
   • Click "Analyze" to see if it's Real or AI.
   • If "Real", switch to the "Identity Risk" tab to see safety metrics.

Tech Stack

• Backend: Python, Flask, PyTorch, Transformers (HuggingFace).
• Frontend: HTML5, CSS3, JavaScript.
• AI Models: ViT (Vision Transformer) based image classifiers.