Remove hidden characters and AI metadata from text and images
Paste text or drop an image. CtrAltRemove finds invisible characters, AI labels and metadata, and the writing patterns that read as machine-made. Then you decide what to remove.
Nothing leaves this page. No upload, no account, no tracking.
Hidden characters in text
Some text carries characters you cannot see: zero-width spaces, odd spaces, direction marks, and a Unicode trick that can hide a whole message. Paste yours to find them. Real emoji and real-language characters are left alone.
AI labels and metadata in images
Many images carry a hidden tag that says how they were made. Platforms read that tag to show a label. This tool shows what is inside a JPG, PNG or WebP and gives you a copy without the tag, location and camera data. The picture itself is not changed.
Writing patterns that read as machine-made
This lists the patterns, quotes the line, and says how to fix it. It does not rewrite your text, give a score, or guess who wrote it. A pattern is a style problem, not proof of anything.
What this cannot do
- It cannot find or remove statistical text watermarks.
- Some AI providers say they shift word choice in a pattern only they can check. There is no hidden character to delete. Rewriting the text in your own words lowers that signal. Nothing guarantees it is gone.
- It cannot remove invisible pixel watermarks from images.
- Tools such as Google's SynthID change the pixels themselves. Stripping metadata does not touch that.
- It cannot tell you whether something was made by AI.
- "No label found" means only that no label was found. It is not proof of how a file was made. AI text detectors are known to give wrong answers, so this tool does not give an AI percentage.
- Use it on your own work.
- Removing a label does not remove your duty to follow platform rules, advertising disclosure rules, school policies or the law where you post. That part stays with you.
Questions people ask
How do I remove zero-width characters from text?
Paste the text into the Hidden text tab and press Scan. The tool lists each invisible character it finds, such as the zero-width space (U+200B), zero-width joiner, byte order mark and soft hyphen. Then press Remove and copy clean text. Real emoji and real-language characters are left alone.
How do I remove C2PA Content Credentials or AI metadata from an image?
Open the Image labels tab and drop in a JPG, PNG or WebP. The tool finds C2PA Content Credentials, EXIF, XMP, IPTC and PNG text data, strips them in your browser, then scans the new file to show they are gone. The picture itself is not re-drawn, so the pixels stay the same.
Does it remove SynthID or other invisible AI watermarks?
No. SynthID and similar marks live in the pixels of an image or in the word choices of text, not in metadata or hidden characters. No metadata cleaner can remove them, and this tool does not claim to.
Is my text or image uploaded anywhere?
No. Everything runs in your browser on your own device. The page makes no network requests while you use it, has no account, and does not log what you paste or drop in.
Is CtrAltRemove an AI detector?
No. It does not say whether something was written or made by AI. It shows what is actually in the file: hidden characters, metadata labels, and writing patterns that often read as machine-made. Finding nothing does not prove how something was made.
Is it free?
Yes. It is free to use with no signup.
What it looks for
- Zero-width space (U+200B), zero-width joiner and non-joiner, word joiner, byte order mark (U+FEFF): characters that take no space on screen but are still in the text.
- Soft hyphen, odd spaces, left-to-right and right-to-left marks: invisible formatting that can break search, links and form fields.
- C2PA Content Credentials: a signed record of how an image was made or edited, stored inside the file.
- EXIF, XMP and IPTC: camera, location, software and caption data that travels with a photo.
What to do about the things it cannot do
Check what is in an image before you change it
Drop the file on Content Credentials Verify to see any signed history it carries. For Google's invisible watermark, SynthID Detector says whether an image, video or audio file was made with AI from Google or its listed partners. Keep the original file. Only share the cleaned copy.
Text watermarks are not something a tool can strip
Google says its SynthID Text watermark survives cropping, a few changed words and mild paraphrasing, and that it has limits. OpenAI announced in October 2026 that it is adding a similar word-choice watermark to ChatGPT and Codex text in the EU. In its own tests on 400-token passages, swapping 25% of the words for synonyms cut detection from about 92% to 17%, and its detector is limited to approved researchers for now. A 2025 research paper reports that paraphrasing, copy-paste edits and back-translation can weaken detection. No scanner can promise a clean result. What helps is real rewriting: your own facts, examples, names, prices and voice. That also makes the copy better, which is the point of fixing it.
Clean text before it goes into your website
- Scan pasted copy here first. Odd spaces and zero-width characters can break search, form fields and links.
- Paste into your page editor as plain text (Ctrl+Shift+V on Windows, Cmd+Option+Shift+V on Mac) so no hidden formatting rides along.
- Run the Writing tab, then replace every flagged phrase with something only your business could say.
Removing the label may not hide it
Platforms do not rely on one signal. Meta says it reads C2PA and IPTC markers and puts invisible watermarks on images made with its own AI. It also asks people to disclose realistic AI video or audio, and may apply penalties if they do not. YouTube says content with C2PA metadata cannot have its AI label turned off, and that creators who keep skipping disclosure can face removal or lose Partner Programme access.
When your content is realistic and made with AI, saying so is usually the safer move.
Sources: Google SynthID Text limitations · SynGuard paper (arXiv 2508.20228) · OpenAI on EU text provenance · Google on SynthID Detector · Meta on AI labels · YouTube AI disclosure