REVEAL: Robust Evolution of Vision-Language Models for Explainable AI-Video Detection

Name: REVEAL: Robust Evolution of Vision-Language Models for Explainable AI-Video Detection
Uploaded: 2026-10-03T01:50:41.992857+00:00
Duration: 951 s
Description: LAVID is a novel agentic framework for detecting diffusion-generated video using Large Vision-Language Models (LVLMs). It improves detection accuracy by enhancing reasoning through explicit knowledge selection, dynamic structured prompting, and online adaptation. This method significantly boosts F1 scores compared to existing detectors on high-quality datasets.

LAVID is a novel agentic framework for detecting diffusion-generated video using Large Vision-Language Models (LVLMs). It improves detection accuracy by enhancing reasoning through explicit knowledge selection, dynamic structured prompting, and online adaptation. This method significantly boosts F1 scores compared to existing detectors on high-quality datasets.

Open the full episode