What we detect.

Images, video, live calls and audio. Four models built in-house, one for each kind of media, combined into a single verdict with the evidence behind it. Each model is measured on named slices and re-scored whenever a new generator family appears.

ImageProprietary

GaussMass

Reads the statistical fingerprints a generator leaves in an image: sensor-noise residue, frequency artefacts and compression history.

93.89 %
recall on edited synthetic images
VideoProprietary

Influx

Scores video frame by frame and across time, catching what a single frame hides.

Frame and time
lip-sync drift, lighting against motion, identity flicker
Live callsProprietary

Sage

Checks a live camera session while it is happening, for a real and present person on the other side.

In session
presentation attacks and injected streams
AudioComing soon

Timbre

Returns a calibrated likelihood that speech was generated, and refuses audio it cannot judge.

Sub-30 ms
to score a clip

Measured by slice.

A single pooled accuracy hides where a detector fails. Each figure below comes from a named slice, and train and validation sets draw on different generator families. How the numbers are made →

Edited synthetic images93.89 %recallMoves 0.64 pp after benign edits
Open real-portrait guard512 / 512passed as authenticN = 512
Real-stress fixture0 / 24false positivesN = 24, research track
Controlled portraits99.22 %real recallV1.3 research candidate

Published profile

Image scoring time14.07 ms on NPU
Image model size90.24 MB
Parameters8 to 15 million
New generator to re-score24 to 48 h
Scaling with image sizelinear
Speed-up, NPU over CPU5.27×

Generator families.

Detectors fail most often when a new kind of generator appears, so each family is measured as its own slice and never pooled into one accuracy. Slice-level results by model version accompany a pilot. Read the white paper ↗

Image · VideoDiffusionMeasured as its own slice
Image · VideoGANMeasured as its own slice
Image · VideoFlow matchingMeasured as its own slice
Image · VideoFace-swap autoencoderMeasured as its own slice
Re-enactmentvideo
Vocoder familiesvoice

Reference datasets

FaceForensics++2019
Celeb-DF (v2)2020
DFDC2020
Deepfake-Eval-20242025
ASVspoof 52024–26
GenVideo2026

What they catch.

GaussMassDiffusion images

Periodic traces that upsampling and diffusion pipelines leave in the frequency domain.

GaussMassGAN images

Statistical fingerprints of a generator, read from noise residue and texture.

GaussMassEdited synthetics

Generated images that were cropped, compressed or retouched afterwards.

GaussMass · InfluxFace swaps

A face placed on another body, in a still or across a clip.

InfluxLip-sync drift

Speech and mouth movement that fall out of step over time.

InfluxLighting against motion

Light on a face that disagrees with how the subject moves.

InfluxIdentity flicker

A face whose identity shifts between frames or cuts.

SagePresentation attacks

A screen, a print or a mask held up to a live camera.

SageInjected streams

Synthetic video fed into a live session in place of the camera.

TimbreGenerated speech

A calibrated likelihood that a voice was generated, not recorded.