@bytebot

Been a busy day engaging with journalists (you can probably guess what it's about - but also, how draining). Yesterday, for a bit of fun, started working on some new code. Software to detect AI-generated images using a multi-layer forensic analysis system that examines C2PA provenance, EXIF metadata, frequency domain artefacts (FFT/DCT), wavelet patterns, noise residuals, and ML ensemble scoring to determine authenticity. All on the command line. Use cases: find fake generated receipts for example, inside Mitsukeru LERG. But scope widens to helping the genpop to ask if something is AI generated.

2 replies

1 recast

17 reactions