SyntheticTrace¶

Status: Research prototype

SyntheticTrace examines still images and video frames for artifacts associated with AI-generated and synthetically manipulated media.

SyntheticTrace is an open-source forensic triage instrument designed to support examiner review of suspected AI-generated and synthetic imagery. It does not issue final authenticity opinions. It identifies technical features, inconsistencies, or artifacts that may warrant further analysis.

GitHub Repository

What It Examines¶

• DCT coefficient distributions for quantisation patterns inconsistent with single-generation capture
• Frequency-domain (FFT) spectra for periodic peaks associated with GAN upsampling architectures
• PRNU sensor-noise residuals for disruption at potential manipulation boundaries
• Related artifact signals combined across a configurable multi-signal pipeline

What It Produces¶

• A methodology-documented, annotated per-frame analysis report
• DCT, FFT, and PRNU visualisations
• A technical finding summary with relevant artifact tables
• A methodology appendix with embedded research citations
• Tool version, parameters, and analysis timestamp, with JSON output for downstream processing

Scope Limitation¶

This tool supports forensic triage and examiner-led analysis. A finding is not, by itself, an authentication opinion. A clean result does not prove authenticity, and a flagged result does not prove manipulation.

Availability¶

GitHub: github.com/ramikhashmel/SyntheticTrace

Install (from source):

pip install git+https://github.com/ramikhashmel/SyntheticTrace.git

Usage:

synthetictrace --input evidence.mp4 --output report/