DeepFake Forensics AI: A Multi-Modal Detection and Blockchain-Anchored Evidence Management Platform

Authors: Naisha Minnah

Published: 2026-05-28 04:47:28+00:00

Comment: 5 pages, 5 figures, 3 tables

AI Summary

This paper introduces DeepFake Forensics AI, a unified platform designed to detect AI-generated synthetic media across image, video, and audio modalities. The system also identifies generative architecture fingerprints and immutably anchors forensic evidence on the Ethereum blockchain, addressing the critical need for tamper-proof evidence preservation in legal contexts. It combines multiple neural network detectors with a blockchain-based chain-of-custody management system, a novel contribution to AI forensics.

Abstract

The proliferation of AI-generated synthetic media poses a critical threat to the integrity of digital evidence in legal and forensic contexts. Existing deepfake detection systems typically address a single modality and provide no mechanism for tamper-proof evidence preservation. We present DeepFake Forensics AI, a unified platform that detects synthetic media across image, video, and audio modalities, identifies generative architecture fingerprints, and anchors forensic evidence immutably on the Ethereum blockchain. Our system trains four independent neural networks from scratch: an EfficientNet-B4 image detector (AUC = 0.9868), a Bidirectional LSTM video detector (AUC= 0.9628), an ECAPA-TDNN audio detector (EER = 18.63%), and a novel GAN fingerprinting module (accuracy = 99.88%) that identifies the generative architecture behind a fake image. Evidence files are hashed with SHA-256, stored on IPFS via Pinata, and registered on-chain via a Solidity smart contract with role-based access control. The platform provides a React frontend and FastAPI backend suitable for deployment in forensic and legal workflows. To our knowledge, this is the first system to unify multi-modal deepfake detection with blockchain-based chain-of custody management.

Key findings

The image detector achieved an AUC of 0.9868 (92.57% accuracy) on FaceForensics++, while the video detector achieved an AUC of 0.9628 (95.54% accuracy) on Celeb-DF v2. The ECAPA-TDNN audio detector yielded an EER of 18.63% (87.99% accuracy) on ASVspoof2019 LA, and the GAN fingerprinting module reached 99.88% accuracy for identifying generative architectures on GenImage.

Approach

The platform employs four independent neural networks: an EfficientNet-B4 for image detection, a Bidirectional LSTM (fed EfficientNet-B4 embeddings) for video detection, an ECAPA-TDNN for audio detection, and a custom residual CNN for GAN fingerprinting. For evidence management, files are SHA-256 hashed, stored on IPFS via Pinata, and registered on the Ethereum blockchain using a Solidity smart contract with role-based access control.

Datasets

FaceForensics++, Celeb-DF v2, ASVspoof2019 LA, GenImage

Model(s)

ECAPA-TDNN

Author countries

India

← Previous