Four pieces of media. Four honest results.
Walk through saved NOCAP analyses of a real photo that looks AI, a known AI-generated image, a deliberately difficult AI-generated video where NOCAP chose to say “I don’t know,” and an AI clip whose signals came back genuinely mixed.
What you’re looking at
The verdict
One of five outcomes — REAL, MIXED, CAP DETECTED, UNCERTAIN, or UNKNOWN. Each is a statement about the observed evidence, never a claim of proof.
The evidence
Individual signals — metadata, structure, compression, model scores — each marked as supporting, suspicious, neutral, or inconclusive. You see what the system saw.
The limitations
Every result states what it cannot tell you. Confidence values are indicative, uncalibrated scores — not probabilities, and never certainty.
The uncertain video sample is included on purpose: it is genuinely AI-generated, and NOCAP abstained rather than guess against conflicting evidence. Abstention helps prevent weak evidence from being presented as certainty.