Paste Claude-generated text below. Our tool rewrites it to break the watermark pattern while preserving your meaning.
Claude's watermark is embedded in the statistical pattern of word choices. It's not a hidden character or metadata, it's woven into which synonyms and phrasings the model selects. To remove it, the text needs to be intelligently rewritten so that ~70% of the original token sequences are disrupted.
Our remover strips hidden Unicode characters and normalizes the text to NFC, without changing your wording. It does not rewrite word choice, so it will not defeat a statistical token-level watermark. For images, it removes C2PA and other metadata.
The invisible statistical pattern in word choices that Claude embeds at the token level. Requires rewriting to remove.
Content Credentials metadata in images and files. Removed by format conversion or re-saving.
Hidden characters can also ride along inside code comments or strings. The cleaner strips them without touching your logic.
There are exactly two real operations here, and it helps to know which you need. Stripping removes hidden and zero-width characters and normalizes the text to Unicode NFC form. It does not change a single word you wrote, so meaning and style are untouched, use it when you just want a clean, safe copy. Rewriting is the heavier option: it paraphrases the text with a different model so the word choices change. Because a statistical watermark lives in word choice, rewriting is the only established way to reduce that signal. The trade-off is that your wording changes, and, importantly, it reduces the signal but cannot guarantee a text will pass every AI detector.
For images, "removal" means stripping metadata: EXIF, XMP, IPTC, and C2PA Content Credentials. Our image tool re-encodes the file without those blocks and hands you a clean copy to download. It does not alter pixels, so it will not remove a pixel-level or invisible image watermark, only the metadata layer, which is where C2PA provenance lives.
No tool can promise your text is undetectable, and we will not claim it. If you are removing a watermark from your own content for privacy or republishing, that is a normal thing to do. If you are trying to pass AI-written work off as human where disclosure is expected, that is on you, the tool is honest even if the use is not.