MAYAVIHIN Deepfake Detection System Version 19

Overview

MAYAVIHIN is a web-based deepfake detection and verification system that analyzes uploaded media files (images, videos, audio) to determine their authenticity using multiple AI analysis agents.

Core Features

Media Upload and Analysis

• Users can upload image, video, or audio files for analysis
• Support for common formats (JPG, PNG, MP4, MP3, WAV)
• File size limits and format validation
• Progress tracking during upload and analysis

Multi-Agent Detection System

The system employs five specialized analysis agents that work in parallel:

1. Visual Forensics Agent - Analyzes images and videos for visual manipulation artifacts
2. Audio Analysis Agent - Detects audio deepfakes and voice cloning artifacts
3. Metadata Agent - Examines file metadata and compression signatures
4. Consistency Agent - Cross-validates results from other agents
5. Explainability Agent - Generates human-readable explanations for detection results

Analysis Results Dashboard

• Overall authenticity verdict (Authentic/Fake/Uncertain)
• Confidence score (0-100%)
• Detailed per-agent results display: Each of the five agents shows individual verdicts (“Real”, “Fake”, “Uncertain”) with corresponding confidence percentages
• Rich explainable AI layout: Visual Forensics, Audio Analysis, Metadata, Consistency, and Explainability agents each display detailed sub-results with human-readable reasoning text and percentage confidence bars
• Multi-agent verdict visualization: Clear display of each agent’s analysis outcome with explanatory text and visual confidence indicators
• Visual indicators highlighting suspicious areas (for images/videos)
• Timeline view for video analysis results

Analysis History

• Users can view their previous analysis requests
• Search and filter past analyses by date, file type, or result
• Export analysis reports

System Logs

• Comprehensive activity logging displaying application events
• Log entries for media uploads, analysis actions, deletions, and system errors
• Scrollable card-based or table layout with columns for timestamp, action type, file name/ID, status, and message details
• Filter and search functionality for quickly locating specific log entries
• Real-time log updates and chronological ordering

Backend Data Storage

• Analysis Records: Store analysis results, confidence scores, agent outputs, and explanations with detailed per-agent verdict data including individual agent confidence scores, reasoning text, and sub-analysis results
• File Metadata: Store original filename, file type, size, upload timestamp
• User Sessions: Track user uploads and analysis history (no authentication required)
• System Logs: Store comprehensive activity logs including timestamps, action types, file references, status codes, and detailed messages

Backend Operations