RemoveClaudeWatermark

C2PA Metadata Remover

Claude attaches signed C2PA provenance metadata to files it generates. Unlike the text watermark, this one genuinely can be removed: re-encoding the image discards every metadata container. Drop files below — the work happens on your device.

Files never leave your device.

What C2PA metadata is

C2PA is an open provenance standard from the Coalition for Content Provenance and Authenticity, used across the industry to record where a file came from. Anthropic attaches a cryptographically signed C2PA manifest to supported file types Claude generates, such as .png, .jpg and .svg. The manifest signals the file was processed by Claude and allows tampering to be detected.

Why re-encoding removes it

C2PA lives in a metadata container alongside the image data, not in the pixels. Decoding an image to a canvas and re-encoding it keeps only the decoded pixels, so the manifest, EXIF and XMP are all dropped. Anthropic's own documentation lists exactly this — "format conversion, re-saving, screenshots, or other means" — among the reasons a file may carry no detectable mark.

This does not apply to text

Claude's text watermark is a statistical pattern in word choice, not metadata attached to a file. There is no container to strip, so no tool of this kind can remove it — including this one. If your concern is generated text rather than generated images, read the Claude watermark guide before assuming any cleaner has helped.

A note on signed provenance

C2PA exists so that people can verify where media came from, which is a genuinely useful thing in a world of synthetic images. Stripping it is legitimate for privacy reasons — the same re-encode removes GPS coordinates and camera serial numbers from your own photos, and every major social platform does it on upload. Passing off a generated image as an unaided photograph is a different matter, and the metadata is not what makes that wrong.

Frequently asked questions

+ Does removing C2PA metadata change the image?

Choosing PNG keeps the pixels lossless and identical. JPEG and WebP re-compress, so there is a small quality cost — use PNG if fidelity matters. Transparency is preserved for PNG and WebP but flattened onto white for JPEG, which has no alpha channel.

+ Are my images uploaded anywhere?

No. The file is read into your browser's memory, decoded to a canvas and re-encoded locally. There is no server involved and no network request — the tool works with your connection off.

+ Does this remove the Claude text watermark too?

No. The text watermark is embedded in the model's word choice rather than in file metadata, so there is nothing for a metadata stripper to remove. Only substantial rewriting, paraphrasing or translation affects it.

+ Will a screenshot do the same thing?

Yes — a screenshot captures pixels only, so all metadata is lost. This tool does the same without the quality loss of screen-resolution capture, and handles batches at once.

More free tools

Related guides