This project aims to build a real-time, multimodal deepfake detection system integrated with blockchain-based media provenance tracking. It is designed to verify the authenticity of digital content—images, videos, and audio—using AI-based forensic analysis and ensure tamper-proof traceability.
Developed under the Safe & Trusted AI initiative, the system addresses the growing threat of misinformation in India and supports fair, explainable, and secure AI deployment.
- Detect manipulated digital content using AI models trained on images, audio, and metadata.
- Provide real-time inference and explainable AI (XAI) outputs to support trust and transparency.
- Ensure media authenticity using blockchain hashing and smart contracts for provenance tracking.
- Design for scalability, multilingual support, and low-resource settings across Indian regions.
- Multimodal Detection Engine: Combines CNNs, Vision Transformers, and GAN detectors for media analysis.
- Forensic Layer: Detects inconsistencies in lip-sync, blink rates, metadata (e.g., timestamps, camera info).
- XAI Module: Generates Grad-CAM heatmaps and attention maps to explain AI predictions.
- Blockchain Provenance Tracker: Stores media hashes and forensic fingerprints using smart contracts.
- User Interfaces: REST APIs, browser plugin, and mobile app for content verification.
| Layer | Tools/Tech Used |
|---|---|
| AI/ML | Python, PyTorch, TensorFlow, Transformers |
| Forensics | OpenCV, Dlib, Librosa, FFMPEG, EXIF Parser |
| Blockchain | Solidity, IPFS, Ethereum Testnet/Polygon, Web3.js |
| Backend APIs | Flask/FastAPI, MongoDB, JWT |
| Frontend | React.js (Web Dashboard), Flutter (Mobile) |
| DevOps/Infra | Docker, AWS EC2, S3, GCP, GitHub Actions |
- FaceForensics++
- Celeb-DF
- DFD (Google DeepFake Detection)
- Custom Indian datasets from news archives, fact-checkers, and regional media partners
- Bias Mitigation: Includes diverse regional, gender, and language data.
- Privacy First: Anonymizes personally identifiable metadata.
- Explainability: Offers visual explanations of AI predictions.
- Open Access: APIs and tools built for public institutions and verified users.
- Phase 1: Prototype & internal testing
- Phase 2: Pilot testing with partners (NDTV, PIB Fact Check, CERT-In)
- Phase 3: Public launch via API, browser plugin, and mobile app
- Scalability: Modular APIs and cloud-based architecture enable national-level deployment.
- Journalism: Verifying user-submitted media and preventing disinformation
- Law Enforcement: Validating digital evidence in legal cases
- Fact-Checkers: Automated backend screening for fake content
- Citizens: Browser/mobile app to check content authenticity instantly