LAVID is a novel agentic framework that leverages Large Vision Language Models to detect videos generated by diffusion models with high accuracy.

• Addresses a critical security gap in AI-generated content detection for videos
• Implements a step-by-step reasoning approach that mimics how humans identify fake videos
• Achieves superior detection performance across various video generation models
• Introduces a comprehensive benchmark for evaluating AI-generated video detection systems

Security Implications: As AI video generation becomes more sophisticated, reliable detection systems are crucial for preserving digital integrity, preventing misinformation, and protecting against privacy violations.

LAVID: An Agentic LVLM Framework for Diffusion-Generated Video Detection