Vastavik

Deepfake Detection Engine connecting…

Access

This instance requires an access token. It is held in this browser tab only and is cleared when the tab closes.

Drop media here, or click to choose

Images, video and audio. Several files at once are fine — they are queued and analysed one after another.

Results

FileVerdictConfidence What was foundTypeTime
—
Confidence —

What was checked

Developer details — raw verdict, scores, coverage, dossier

Why

    AI signature check

      Artifact family coverage

      FamilyStatusScore
      Full dossier (JSON)

      Teach the system

      Authorised reviewers only. Tell the engine whether the verdict above was right. Your answer is recorded against your name and reviewed before it influences any model — submitting feedback does not retrain anything.

      How to read this. Deepfake detected means manipulation signatures were found and passed the engine's accusation threshold. No deepfake detected means the checks that ran found nothing — it is the absence of evidence, not proof a file is genuine, which is why the checks that ran are listed. Possibly manipulated means signals were present but fell short of the threshold. Authentic appears only when cryptographic content credentials prove origin, so it is deliberately rare. The engine is fail-closed: it would rather say it does not know than guess.

      How the verdict is reached. Almost always from the media itself — the pixels or audio samples. The one exception is a file that declares its own synthetic origin, such as C2PA content credentials naming the model that made it; that is conclusive and the verdict says so. The absence of such a marker means nothing, because messaging apps strip them routinely, which is why the media analysis has to stand on its own. The full signature check is in Developer details.