The verification toolbox.
Four models, one evidence layer. Each is built for a different kind of media and a different kind of attack; together they give a reviewer a verdict and the reasons behind it.
GaussMass
Forensic image analysis with industry-leading accuracy.
GaussMass reads the statistical fingerprints a generator leaves behind: sensor-noise residue, frequency-domain artefacts, compression history, and the small inconsistencies a synthetic pipeline cannot help introducing. It returns a verdict with the evidence attached, in real time.
- Real-time detection
- Image deepfake detection
- Real-time analysis API
- Enterprise-grade security
Influx
Next-generation video detection with temporal analysis.
Influx processes video frame by frame and across time, catching the inconsistencies that a single frame hides: lip-sync drift, lighting that disagrees with motion, identity that flickers between cuts. It handles high-resolution streams with minimal latency.
- Real-time detection
- Temporal consistency
- High-res stream support
- Minimal latency
Sage
Liveness and deepfake detection for live camera feeds.
Sage performs real-time biometric verification on live video, detecting presentation attacks and injected synthetic content while the session is happening. Multi-factor liveness checks confirm that a real, present human is on the other side of the camera.
- Real-time biometrics
- Spoof-attack detection
- Injection detection
- Multi-factor liveness
Timbre
Calibrated synthetic-speech evidence, with the error rates beside every verdict.
Timbre scores an audio clip and returns a calibrated likelihood that the speech was generated rather than recorded, alongside the conditions and error rates it was measured under. Audio it cannot judge, under a second, near-silent, or below an 8 kHz source rate, it refuses rather than guesses on, because a number that should not exist will still get used.
- Calibrated likelihood
- Error rates per verdict
- Refuses unjudgeable audio
- Sub-30 ms scoring