// dfd // multi-modal authenticitylatest: 2026-06-26 · Audio detection arrives →

IS THIS REAL?

AI generators leave detectable patterns in the files they produce — patterns different from how real cameras, microphones, and editing tools encode their output. Drop a video, photo, or audio clip here and we compare its fingerprint against a curated corpus per modality.

live

0last hour

0last 24h

0today

72all-time

── module / fleet stats

ALL-TIME ANALYSES

submissions

by modality

VIDEO41

IMAGE25

AUDIO6

by verdict

FAKE47

REAL24

INCONCLUSIVE1

ai generators caught

udio×1

── module / active rulesets

DETECTION VERSIONS

which model state is scoring submissions right now, per modality.

modality	version	derived
video	`v.b438e85c`	13d ago(promotion)
image	`v.634311f6`	25d ago(verified upload)
audio	`v.2ab602a5`	56d ago(verified upload)

// step_01

Extract

Metadata fingerprint per modality — H.264/x264 encoder options for video, EXIF / ICC / quantization tables for images, ID3 / codec details for audio. Direct AI-generator signatures (Midjourney, ElevenLabs, Sora, …) get flagged immediately if present.

// step_02

Compare

Each field is checked against statistical thresholds derived from the real vs fake corpus for that modality. Mismatches accumulate into a verdict.

// step_03

Verdict

FAKE / REAL / INCONCLUSIVE. Every rule that triggered is shown so you can judge the reasoning — this is a transparent detector, not a black box.