Frequently Asked Questions

Our scanner (v18) uses a three-layer approach: SpecTTTra forensic analysis checks for AI-model-specific spectral fingerprints; the Fakeprint layer cross-references known Suno/Udio generation artifacts; and a Consistency check evaluates temporal patterns that differ between AI-generated and human-performed audio. Results are combined into a single Human / AI probability score.

Every detector uses different features and training data. Our system is tuned specifically for AI music models (Suno v3–v4, Udio). General-purpose detectors trained on speech or older models will give different results. A 'Human' score on one tool and 'AI' on another is completely normal — there is no universal ground truth.

False positives happen, especially on highly produced tracks with drum machines, heavily auto-tuned vocals, or very clean studio recordings. Our system is tuned for near-zero false positives on human music (0% in internal testing), but edge cases exist. If you believe your result is wrong, use the Contact form and we'll review it.

Newer model versions (v5+) produce significantly different artifacts than the training data our detector was built on. We openly disclose this — our scanner catches ~66% of Suno v5 and ~3% of Udio v2 tracks. We're actively working on v8 detector research for the next generation. Honesty matters more to us than inflated claims.

No. Scanner results are computed in your browser session and are not stored on our servers. We do not share any audio content or analysis results with third parties.

The 'Before' scan analyzes your original uploaded file. The 'After' scan analyzes the mastered output — useful for seeing how the Anti-AI preset changes the detection score. For the best evasion result, run the Anti-AI preset and check the After score.