C2PA · EXIF · documents
File marks live in the container. Not in the picture you see.
A photo or a PDF can carry a signed Content Credential, a camera diary, or an author field. Those are bytes with names. Gifi can list them, then delete the ones that actually sit in the file.
C2PA Content Credentials, EXIF, XMP and document properties are metadata. They are not Claude’s SynthID-Text trail, and they are not a watermark printed into the pixels. Stripping them removes the local record. Soft binding can still re-link a file to a remote manifest. Pixel-embedded image watermarks, audio and video are out of scope.
Two jobs, because the marks are different
Claude’s text watermark is a pattern in word choice. There is nothing to delete. A rewrite is the only honest lever, and it will change the prose.
A file credential is the opposite. C2PA writes a signed note into metadata. Cameras write EXIF. Editors write XMP. Word and PDF keep author fields. Those can be stripped without touching the words on the page or the pixels in the photograph — unless the mark was printed into the appearance itself, which we do not handle.
What the cleaner actually deletes
Images — PNG, JPEG, WebP — lose EXIF (including GPS), XMP, C2PA manifests and comments. ICC colour profiles stay, so colours do not shift. The pixels are not re-encoded.
DOCX and ODT lose author and company fields, custom properties, revision-save IDs, comment authors, the page-one thumbnail, and metadata inside embedded images. The title is kept. The body is not rewritten.
PDFs lose the info dictionary and XMP, then the file is rebuilt so leftover objects are not recoverable. That job goes to a worker. A format we cannot fully clean is refused rather than half-stripped.
Soft binding, and other things a strip cannot promise
Some C2PA deployments keep a fingerprint and a copy of the manifest on a remote service. After the local label is gone, that fingerprint can still match. The industry name is soft binding. We can promise the embedded credential is gone. We cannot promise a registry has forgotten the file.
Removing a note does not change how the pixels were made. It does not prove a human took the photograph. It does not disturb a statistical watermark in a caption sitting next to the image.
What a file strip does, and does not, mean
- What does the cleaner remove from a photo?
- EXIF (including GPS), XMP, C2PA manifests, and comments. ICC colour profiles stay, so colours do not shift. PNG, JPEG and WebP are stripped losslessly — the pixels are not re-encoded.
- Does stripping C2PA prove an image was not AI-made?
- No. Removing a signed note removes the note. It does not change how the pixels were made, and it does not prove human authorship.
- What is C2PA soft binding?
- A fingerprint that can re-link a file to a remote manifest after the embedded credential is gone. A metadata strip deletes the local label. It cannot promise the remote record is gone.
- What happens to a PDF?
- Document info and XMP are cleared, then the file is structurally rebuilt. Without that rebuild the old metadata stays recoverable in the bytes. PDFs go to a worker that can do that job properly.
- What about Word and OpenDocument files?
- Author and company fields, custom properties, revision-save IDs, comment authors, the page-one thumbnail, and metadata inside embedded images. The title is kept. The document body is not rewritten.
- Which formats can Gifi clean?
- PNG, JPEG, WebP, PDF, DOCX and ODT. A format we cannot fully clean is refused rather than half-stripped. Pixel-embedded image watermarks, audio and video are out of scope.
Choose the right tool
Metadata in a file? Strip it. A trail in the words? Rewrite.
Drop a file on the home page to see what is riding along before you spend a credit.