AI Text Watermark Removal FAQ

A hidden signal embedded in text so its origin can be identified later. It takes three forms: invisible Unicode characters inserted between words, lookalike characters swapped for ordinary ones, and a statistical bias in the model's token choices during generation. The first two live in the bytes and can be found by inspection; the third cannot.

The process of detecting and deleting those hidden markers. In practice it means two jobs: stripping invisible Unicode characters and homoglyphs from the text itself, and removing provenance metadata such as C2PA manifests, EXIF and XMP blocks from image and document files.

Paste your text into the tool on the homepage and press Clean text. Every invisible carrier is deleted and homoglyph spaces are normalized, and you get a report of exactly what changed. For files, use the Files tab and download the cleaned copy.

Yes. No account, no trial, no usage cap for normal use, and no watermark on the output. Uploaded files are limited to 25 MiB each.

Zero-width spaces, joiners and non-joiners, word joiners, the byte-order mark, soft hyphens, bidirectional controls, variation selectors and Unicode tag characters are deleted. Space homoglyphs such as non-breaking, narrow, thin, hair and ideographic spaces are normalized to a standard U+0020 space.

C2PA manifests, EXIF and XMP metadata blocks, and known AI-generator markers, from PNG, JPEG, SVG, PDF, DOCX, ODT, HTML and Markdown. The visible content (pixels, text, layout) is preserved exactly.

No. The text cleaner never paraphrases, reorders or rewrites; it only deletes characters with no visible glyph and normalizes lookalike whitespace. The file cleaner leaves pixel and content streams untouched and only rewrites the container's metadata segments.

Yes, for everything it claims to do. Run Inspect only on the cleaned text and it should report zero suspicious characters; re-upload a cleaned file and its metadata report should come back empty. Anything the tool could not parse is reported rather than silently skipped.

Not in the current interface. It handles one file per upload. Batch and ZIP handling is on the roadmap, and the API endpoints can be scripted in the meantime.

No. Those tools score writing style statistically and ignore invisible characters entirely, so a cleaned document scores exactly the same as the original. Any service claiming otherwise is misrepresenting how detection works. This site does not offer detector evasion.

No, and neither can anyone else by inspecting the file. Statistical watermarking biases token selection during generation and leaves no special characters behind, so there is nothing in the bytes to strip. It degrades under heavy paraphrasing and translation, but that is a rewrite, not a removal.

No. Visible watermarks, logos and bylines are part of the image content, not hidden metadata. This tool only removes metadata and invisible characters.

It depends on the content and your jurisdiction. Cleaning your own drafts or content you are authorized to process is ordinary hygiene work. Stripping provenance to misrepresent authorship, evade an academic integrity policy, or defeat a legal disclosure requirement is not, and is outside our acceptable use terms.

No. Input is processed in memory for the duration of the request and then discarded. Nothing is written to disk, nothing is logged as content, and nothing is shared with a third party.

No. There is no sign-up, no login and no email required.

Yes, 25 MiB per uploaded file. Text pasted into the Text tab has no practical limit beyond what your browser can handle.