Hidden characters in the original
Clean text
Images
EXIF, XMP, IPTC, comments, C2PA / Content Credentials and embedded prompts (Stable Diffusion, ComfyUI). Lossless, colour profile kept.
Documents
Author, company, generating app, template and custom properties in Office & OpenDocument files, plus their embedded images.
Info dictionary (Title, Author, Creator, Producer…), XMP metadata for the document and each page, and attached C2PA manifests.
Text
Zero-width and special spaces, bidi controls, hidden Unicode tags, steganographic selectors and chat leftovers like oaicite.
What it can’t remove: invisible watermarks embedded in the pixels themselves or in the statistical pattern of the words (e.g. SynthID). Those aren’t metadata — they’re part of the content.