Pipeline
How MetaClean reads your bytes
Metadata is not magic — it is structured data parked in predictable corners of the file. MetaClean walks those corners directly.
Detection
- JPEG — scans APP markers: APP1 (EXIF/XMP), APP13 (IPTC/Photoshop), APP2 (ICC), COM comments, APP11/JUMBF for C2PA. A minimal TIFF/EXIF walk extracts GPS coordinates when present.
- PNG — walks chunks:
tEXt,zTXt,iTXttext chunks,eXIf, plus a C2PA/JUMBF signature scan. - WebP — walks RIFF chunks:
EXIF,XMP,ICCP, C2PA scan.
Cleaning
Once you approve, the image is decoded and re-encoded through the Canvas API. The output contains only pixel data and the format's required structural chunks — every metadata container is gone by construction, not by selective deletion that could miss an edge case.
What it deliberately does not do
Pixel-level edits. Visible watermarks, AI-detection patterns baked into the image data, and steganography are not metadata and are out of scope. MetaClean is honest about that boundary so you know exactly what you are getting.