AI-Based Deepfake Video Detection Tool

This project is a comprehensive, modular system for detecting deepfakes in both audio and video using state-of-the-art AI models. Its architecture separates concerns across dedicated components, enabling scalable deployment and easy integration into real-world applications, with a focus on forensic use for law enforcement.

Project Structure

• ai (Python): Deep learning models and inference scripts.
• backend (Express.js): API layer and communication management.
• frontend (Next.js): User interface for interacting with the tool.

---

💻 Installation & Running

Each component is self-contained and requires its own setup.

Frontend (Next.js)

1. Navigate to the frontend directory:

   cd frontend

2. Install dependencies:

   npm install

3. Start the development server:

   npm run dev

Backend (Express.js)

1. Navigate to the backend directory:

   cd backend

2. Install dependencies:

   npm install

3. Start the backend server (port 5000):

   node server.js

AI (Python)

Note: This service also runs on port 5000. Run it independently or use a port management tool if needed.

1. Navigate to the AI directory:

   cd ai/rushi

2. Install required Python libraries:

   pip install -r requirements.txt

3. Run the AI service:

   python app.py

---

🤖 AI Models

Located in ai/model:

• best_deepfake_detector.pth: Detects deepfakes in audio files.
• best_optimized_fine_tuned_model.h5: Detects deepfakes in video files.

---

🧠 Problem Statement & Analysis

AI-generated media (deepfakes) are increasingly realistic, posing risks like misinformation, identity theft, and erosion of public trust. This project provides an AI-powered forensic tool for law enforcement to detect, analyze, and securely document deepfake videos for reliable evidence gathering and court presentation.

---

🛠️ Key Features

1. Multi-Modal Detection

   • Facial Manipulations: Detects subtle anomalies and inconsistencies in facial features and movements across common video formats.
   • Voice Cloning: Analyzes audio tracks for voice manipulations and synthetic speech.

2. Detailed Forensic Analysis

   • Binary Classification: Classifies videos as "Real" or "Deepfake".
   • Confidence Scores: Provides a numerical confidence score for each classification.
   • Visual Heat Maps: Highlights manipulated areas in video frames.

3. Tamper-Proof Reporting

   • Automated PDF Generation: Creates comprehensive, tamper-proof PDF reports.
   • Report Contents: Includes metadata, cryptographic hashes, detection results, and key snapshots with heat map overlays.

4. Secure Chain of Custody

   • Secure Storage: Ensures secure storage of media and reports.
   • Blockchain-based Audit Trails: Uses a private blockchain ledger for immutable audit trails, guaranteeing forensic integrity.

---

⚖️ Expected Outcome

This tool provides law enforcement with a reliable, efficient, and legally defensible forensic solution for deepfake detection. It enables quick identification of manipulated media and validated reporting for court use, helping combat the threat of deepfakes and digital misinformation.