About DetectVideo

A clearer first read on suspicious video

DetectVideo estimates whether a file shows signs of AI generation or manipulation by reviewing visual, temporal, audio, metadata, and provenance signals together. It is built for teams that need a fast, defensible assessment before they publish, escalate, moderate, or rely on a clip.

Use cases

Who it's for

DetectVideo is most useful when someone needs to decide what to trust, what to escalate, and what needs deeper review.

Editorial review

Newsrooms

Pressure-test eyewitness clips, source submissions, and viral footage before publication.

Platform response

Trust & Safety

Triage suspicious uploads, impersonation attempts, and synthetic content campaigns without treating every clip like a full forensic case.

Case intake

Investigations & Legal

Add structured signal review to evidence intake, internal investigations, and case preparation.

Signal map

What we analyze

Every report is grounded in multiple signal types, not a single visual guess.

5 signal groups
Frames

Visual signals

Frame artifacts, lighting inconsistencies, face-detail instability, and compression patterns.

Continuity

Temporal signals

Flicker, motion drift, scene continuity, and lip-sync behavior across frames.

Speech

Audio signals

Speech cadence, spectral artifacts, and voice patterns that may suggest synthesis.

File trail

Metadata signals

Codec details, timestamps, export traces, and other file-level clues.

Origin

Provenance signals

Available origin evidence and packaging history that help explain where a file came from.

Honesty note

What the result means

DetectVideo returns AI-likelihood assessments, not certificates of authenticity. The output reflects the strength and consistency of the signals available in the file.

Short clips, heavy compression, screen recordings, re-exports, and missing audio can reduce certainty. A low AI-likelihood result does not prove a video is real, and a high result should be reviewed alongside source context, provenance, and human judgment.

Use source context
Check provenance
Keep human review