AiEraser scans an image for every metadata block it carries — EXIF, IPTC, XMP, ICC, and C2PA/JUMBF content-credential manifests — shows you exactly what's inside before touching anything, then strips it all in your browser. Nothing is uploaded for JPEG, PNG, or WebP files.
Only the pixels. AiEraser re-encodes the image through your browser's Canvas API, which by construction keeps the visual content and drops every metadata block — there's no selective "keep this field" option because the removal method itself doesn't distinguish between fields once you've chosen to clean a file.
Metadata removal doesn't touch anything baked into the pixels themselves. Invisible watermarking schemes (like Google's SynthID) and visual AI-detection classifiers operate on pixel data, not metadata, and survive this process entirely. If a platform's AI label is driven by a visual classifier rather than metadata, removing the manifest won't change the outcome.