unmarking

Frequently asked questions

What content credentials are, what unmarking removes and what it leaves alone, and how to check the result.

What is a C2PA content credential?

A signed manifest, defined by the Coalition for Content Provenance and Authenticity (C2PA) and also known as Content Credentials, that is stored inside a file next to the media data. It records which tool produced the file and what was done to it since. Claude, ChatGPT, Adobe Firefly and other AI tools attach one to the images, video and audio they generate.

Does removing the credential change the image, video or audio?

No. The manifest is a separate block in the file container. unmarking cuts that block out and leaves the pixels, frames and audio samples byte for byte as they were. Other metadata such as EXIF is kept, so the only difference is that the file gets smaller by the size of the manifest.

Does unmarking remove SynthID or other invisible watermarks?

No. Invisible watermarks such as SynthID are embedded in the pixels or audio samples themselves and are a different mechanism from content credentials. unmarking does not touch them, and it does not remove watermarks in generated text either.

Which file formats are supported?

Images: JPEG, PNG, GIF, WebP, TIFF, DNG, HEIC, HEIF, AVIF and SVG. Video: MP4, MOV and AVI. Audio: M4A, MP3 and WAV. The API and MCP server additionally accept JXL and FLAC. Files can be up to 256 MB each. PDF is not supported.

Is anything uploaded?

Not from the website. The file is read, rewritten and verified in your browser with WebAssembly, and the page makes no analytics or third-party requests. Only the HTTP API and the MCP server receive files, and they process them in memory without storing them.

How can I check that the credential is gone?

unmarking inspects every result before offering it for download, and each cleaned file links to an independent checker: Anthropic's file checker for files that were signed by Claude, and Content Credentials Verify for credentials from other tools. Both should report that the file carries no content credential.

What about files that only link to a credential hosted elsewhere?

Some files carry no manifest of their own and only reference one at a remote URL. unmarking reports these as remote and leaves them unchanged, since the file itself contains nothing to remove.

Can I use unmarking from a script or an AI agent?

Yes. The same engine runs as a free HTTP API and as a hosted MCP server at api.unmark.ing, with no key or account. Both are documented on the developer page.