LADDER: Landmark-Only Lip-Sync Deepfake Detection Across Generators and Languages
BioLip detects deepfakes by analyzing biomechanical constraints in lip motion using only landmark coordinates, ignoring pixels and audio. It finds that synthetic motion lacks physical limits, resulting in unnatural temporal jitter in velocity and acceleration. This kinematic variance serves as a universal signal for forgery across different generators and languages.