AI video detector

A video needs more than one frame.

AI Screener samples a bounded short clip, keeps the representative-frame scores separate from temporal motion, container metadata, audio-track, and provenance evidence, and names missing coverage clearly.

01

Container and codec inspection

FFmpeg and FFprobe inspect safe-to-read streams, duration, codec information, audio presence, and editor metadata without trusting the filename.

02

Representative frame scoring

A pinned open synthetic-image model examines evenly distributed frames, then presents the score timeline rather than hiding a single peak.

03

Temporal consistency

Second-order motion analysis looks for abrupt changes across a low-resolution timeline. It is a clue, not a universal deepfake detector.

04

Partial-result reporting

If a decoder, audio track, or optional provenance check cannot run, the report says so and keeps the remaining evidence visible.

No automated detector can prove authenticity. AI Screener combines independent signals, shows the evidence, and leaves the decision with a human.

Run the free checker

Questions to ask of the result

Limits matter as much as the signals.

Can a video detector prove a clip is human-made?

No. A negative research signal is inconclusive; request the original file or verified provenance for consequential decisions.

What video files can I check?

The public checker accepts one MP4, WebM, or AVI file at a time, with a bounded duration so analysis stays safe and explainable.