Forensic Pipeline

A transparent look at how we distinguish pixels from probability.

1. Secure Ingestion

When you upload an image, it's processed in a secure, ephemeral memory sandbox. We immediately strip potential malicious payloads and normalize the file (handling HEIC, WebP, and RAW formats) into a standardized tensor matrix ready for analysis.

2. Parallel Forensics

The image is analyzed simultaneously by three distinct engines:

  • Visual Artifacts: Scans for unnatural smoothness and pixel grid irregularities common in diffusion models.
  • ELA (Error Level Analysis): Detects compression inconsistencies that indicate Photoshop manipulation.
  • Metadata Scan: Verifies C2PA integrity and camera sensor signatures.

3. Statistical Synthesis

We don't rely on a single score. Our ensemble algorithm weights the inputs from all three engines. A high "Visual Artifact" score might be discounted if the "Metadata" strongly confirms a specific camera model, preventing false positives from heavy JPEG compression.

4. Actionable Verdict

You receive a clear probability score (e.g., "98% AI Generated") along with a downloadable PDF report detailing the specific reasons for the flag.

Try it now