How ZKDetect Works

A comprehensive look at our privacy-preserving deepfake detection process using Zero-Knowledge Proofs.

Step 1

Video Upload & Preprocessing

Your video is securely uploaded and preprocessed locally before any analysis begins.

End-to-end encryption during upload

Local preprocessing to protect privacy

Perceptual hash (pHash) generation

Metadata extraction and sanitization

Step 2

AI Model Analysis

Advanced EfficientNet-ViT hybrid model analyzes the video for deepfake indicators.

EfficientNet feature extraction

Vision Transformer attention analysis

Multi-scale deepfake detection

Confidence score generation

Step 3

Zero-Knowledge Proof Generation

Generate cryptographic proof that validates the AI prediction without revealing sensitive data.

zk-SNARK circuit compilation

Witness generation from AI output

Proof generation with private inputs

Public verification key creation

Step 4

Smart Contract Verification

Submit the proof to Ethereum smart contract for decentralized verification.

Proof submission to blockchain

Smart contract verification

Gas-optimized verification

Immutable result storage

Step 5

Result Processing

Process and format the verification results for user consumption.

Result interpretation

Confidence level calculation

Timestamp verification

Transaction hash generation

Step 6

IPFS Storage

Store metadata and proof data on IPFS for decentralized access and permanence.

Metadata IPFS upload

Content addressing

Distributed storage

Public accessibility

Privacy-First Architecture

Our system ensures that sensitive data like video content, perceptual hashes, and AI prediction scores never leave your device in plaintext. Only the cryptographic proof is shared, maintaining complete privacy while enabling public verification.