Know if an image or video is real.

All weights below are relative and share a single normalized pool — only their ratios matter. Setting them all equal makes every signal contribute the same share; raising one relative to the others emphasizes it.

Adjust how much each detector influences the final decision. Every signal — face, generative, latent reconstruction, spectral, attribution, manipulation, and (for video) voice / audio — is treated identically in one weight-normalized blend. The report shows each signal’s resulting share of the decision, which sums to 100%. Results recompute live.