FaceVid-Forensics-100K dataset introduces multi-agent forensic reasoning for deepfake detection
August 6, 2026
FaceVid-Forensics-100K provides 100,000 videos covering 33 synthesis methods to improve deepfake detection. The framework uses multi-agent forensic reasoning to overcome the limitations of single-perspective MLLMs in capturing subtle forgery artifacts across diverse generation methods.
HOW THIS AFFECTS YOU
●
researcherUse this dataset to train more robust detectors capable of generalizing to new synthesis techniques.
●
policyBetter detection tools are necessary to mitigate the societal risks posed by advanced generative video.