Deepfake Detection Explained

Deepfakes succeed by tricking human visual perception, but they leave mathematical scars in the pixel data. Detection algorithms look for what humans cannot see.

Spectral analysis examines the high-frequency components of an image. Generative Adversarial Networks (GANs) and diffusion models struggle to reproduce the exact frequency distribution of a real camera sensor. They leave a subtle 'checkerboard' artifact in the Fourier domain.

Additionally, temporal inconsistencies in video—such as unnatural blink rates or micro-variations in skin blood-flow (rPPG)—provide high-confidence signals of manipulation. TruthLens applies these filters automatically.