Find the invisible
See the Unicode name, code point, and count for every non-printing formatting character we detect.
Free · private · open source
Inspect AI-generated text for invisible Unicode characters, remove formatting artifacts, and copy a clean version. Everything happens on your device.
An honest distinction. Invisible characters can be removed directly. Statistical watermarks such as SynthID are patterns in token selection, not hidden symbols.
No removable formatting characters detected.
What this tool does
See the Unicode name, code point, and count for every non-printing formatting character we detect.
Your text stays inside the browser. No account, upload, model call, or server-side retention.
Ordinary words and punctuation remain untouched. The tool does not “humanize” or rewrite your ideas.
SynthID, explained plainly
Google DeepMind’s SynthID adjusts token probabilities during generation. Detection compares the resulting pattern with what a configured detector expects. That is fundamentally different from stripping a zero-width space.
Read how AI text watermarks work →Straight answers
The phrase can mean two different things: hidden Unicode formatting added to copied text, or a statistical signal embedded in an AI model’s token choices. Unmark Text finds the first kind.
No generic text cleaner can honestly guarantee that. SynthID changes token probabilities while a model generates text; it is not a layer of invisible characters. Unmark removes inspectable Unicode artifacts and explains the difference.
No. The web cleaner runs locally in your browser. Analytics records feature usage and character counts only, never the text you paste.
The cleaner identifies zero-width spaces and joiners, bidirectional formatting marks, word joiners, soft hyphens, byte order marks, and related non-printing code points.
No black box
Unmark Text is MIT-licensed. The cleaner, API, tests, and technical notes are public.
Explore the repository ↗