Hybrid RAG-Enhanced Deepfake Detection: A Novel Approach Combining Retrieval-Augmented Generation with Visual Inconsistency Analysis

This research proposes a novel hybrid framework for deepfake detection that integrates Retrieval-Augmented Generation (RAG) with visual inconsistency analysis. The system utilizes a dual verification strategy: contextual verification using ViT and FAISS for semantic similarity, and AI-generated artifact detection using YOLO and SAM. It introduces uncertainty quantification and operates entirely offline for privacy, making it suitable for journalism, law, forensics, and personal content verification. This paper aims to contribute a scalable, interpretable, and privacy-preserving system for combating manipulated media in the age of generative AI.