Upload a video or audio file. Clarifake analyzes visual, temporal, and audio signals to estimate whether synthetic generation may be present.
Every upload goes through the same forensic pipeline, whether it's a 10-second clip or a full interview.
Drop in a video , image or audio file, Clarifake extracts frames, waveforms, and metadata automatically.
Multiple independent models check for generation artifacts — spectral noise, lip-sync drift, unnatural blinking, and more.
Signals are combined into a single 0–100 score and a clear verdict: authentic, synthetic, or inconclusive.
You're not getting a black-box guess. Every score is built from four independent checks, and each one is shown to you separately.
Frequency patterns inconsistent with natural camera capture.
Frame-to-frame flicker, warping, and motion drift.
Frame 001 → 002 → 003 → 004 → ⚠ driftMismatches between speech and lip or facial movement.
Encoding history, C2PA credentials, and file provenance.
A score above 85 is flagged as high-confidence synthetic. Between 40–85, Clarifake returns "inconclusive" rather than guess.
Two clips of the same scene — one untouched, one AI-generated. The frame analysis is what tells them apart.
Verify footage before publishing.
Flag suspicious uploads automatically.
Preserve evidence and generate reports.
Check before you believe. Check before you share.
Catch impersonation attempts and cloned likenesses before they do damage.
Support cyber fraud investigations and preserve tamper-evident digital evidence.
The original brief called for detection, verification, tracing, and investigative support. Origin tracing, dissemination mapping, impersonation detection, and evidence-integrity workflows have been built out alongside the core detection engine — nothing here is a partial anymore.
| Requirement from problem statement | Clarifake |
|---|---|
| Detect AI-generated videos/images | ✅ |
| Detect AI-altered media | ✅ |
| Verify authenticity | ✅ |
| Identify manipulation signals | ✅ |
| Impersonation detection | ✅ |
| Cyber fraud investigation | ✅ |
| Digital evidence tampering | ✅ |
| Signal | Clarifake |
|---|---|
| Pixel artifacts | ✓ |
| Temporal consistency | ✓ |
| Audio-visual sync | ✓ |
| Metadata | ✓ |
| Confidence score | ✓ |
| Explainable result | ✓ |
Every scan ends with a verdict, a confidence score, and the underlying evidence — exportable as a forensic report you can attach to a case file or editorial review.
No account needed for your first three checks.