Feature Spotlight

Multimodal
Forensics.

Correlating spectral anomalies across audio, visual, and metadata domains to identify synthetic deceptive media.

What this feature does

Multimodal Forensics is the core detection layer of the Veridex engine. It doesn't look at one signal in isolation. Instead, it audits the 'Forensic Consistency' between different media streams within a single filename. It looks for the minute mathematical artifacts left behind by generative AI models—specifically GANs (Generative Adversarial Networks) and Diffusion-based architectures.

Read Inputs

  • • Pixel-level frequency data
  • • Audio spectral density
  • • Temporal frame-boundary data
  • • EXIF & Internal metadata

Produces Outputs

  • • Probability heatmap
  • • Spectral anomaly report
  • • Synthetic marker verification
  • • Confidence-scored verity index

How to interpret results

A high 'Synthetic Marker Detected' flag indicates that the media track contains frequency arrangements or macro-block inconsistencies that are over 90% correlated with known generative model signatures. This is an indicator of risk, not a binary proof of deception.

What it cannot guarantee

We cannot guarantee the identification of 'Post-Detection Compression' (PDC) artifacts that are intentionally added to media to mask synthetic signatures. Forensic signals degrade with every re-upload or re-encoding.