NUis this AI generated?

Sampled-frame video checker

Check frames.
Not motion.

The browser samples up to five representative frames and submits those images for an aggregate warning signal. The original video stays in the browser.

Open the free checker

Scope before score.

Every result is a warning signal. The evidence and limitations below define what this route can and cannot support.

Browser sampling

H.264 MP4 and WebM work most reliably; short clips use fewer frames and longer clips use up to five.

Aggregate only

The API reports frame count, scored, pending and failed frames plus median, mean and maximum scores.

Existing evidence

Video is judged on 15 sampled frames, at least 3 of which must score strongly: 82.6% caught (62.9-93.0%) on 23 videos at 1.4% false warnings. That interval is wide because the set is small. Sampled video says nothing about motion, lip-sync, or the moments between sampled frames.

Do not infer more than the lane measured.

  • This is not temporal deepfake analysis.
  • It does not evaluate lip-sync, motion continuity or audio-video consistency.
  • Unsampled moments can contain signals the checker never sees.

Continue with evidence

How sampled video worksRead more →MethodologyRead more →Detection APIRead more →