Runs in your browser. Nothing uploaded, no limit.
Free Llama watermark detector: scan text copied from Llama for hidden Unicode characters in your browser. No upload, no account, no limit.
Llama watermark detector
Paste in text copied from Llama and this tool scans it for the zero-width and non-printing Unicode characters a chat interface can leave behind, without changing anything visible. It's built for the question people actually search: does Llama hide a watermark in its output.
The scan runs entirely in your browser. Nothing is uploaded, there's no account, and there's no limit on how much text you check.
How it works
It takes seconds and nothing you paste leaves this page.
Why it matters
A zero-width space inside a heading pasted from Llama splits the keyword, so a slug or anchor stops matching the visible title. A non-breaking space carried over from whichever interface rendered the reply can break a shell command, a config file or a CSV column without any visible sign of why.
The same characters cause broken find-and-replace, failed CMS saves, and layout problems in exported PDFs. None of it requires anything unusual, it's a side effect of flattening formatted text, wherever that formatting happened.
Where they come from
Because Llama's weights are openly available, dozens of unrelated interfaces serve the same underlying model: consumer apps, developer playgrounds, self-hosted tools and API wrappers with their own front ends. Every one of them has its own Markdown renderer and its own way of converting formatted output back to plain text on copy.
That's the single most checkable fact about hidden characters in Llama text: what you find says far more about which specific interface produced the copy than it does about the model underneath it. Two people quoting an identical Llama answer through two different hosts can end up with entirely different invisible residue.
Straight answer
No, and for Llama specifically this matters even more than usual. A hidden character reflects the interface's formatting pipeline, not the model, and Llama has no single interface. The same characters turn up in text copied from Word, Google Docs and countless other tools with nothing to do with Llama at all.
It also isn't evidence about whether text will pass a separate AI detector. Those tools weigh sentence patterns and vocabulary, not invisible bytes. Treat a finding as a formatting lead, nothing more.
Who uses it
Developers building on Llama through an API or a self-hosted deployment, checking output before it's stored or displayed. Writers and researchers pasting a Llama answer, from whichever front end they used, into a document or a CMS where a stray character could split a slug or heading.
Anyone comparing Llama output across two different hosting providers, where a hidden-character difference is a clue about the interface, not the model, runs this to sort out which is which.
FAQ
Everything below is answered in full, nothing to click open.
Not in the sense of a deliberate hidden signal. Text copied from Llama can carry ordinary invisible characters, and because Llama runs through so many different interfaces, which characters show up depends heavily on which one you used.
Paste it into the box above and press Scan text. Every hidden character is listed with where it sits.
Yes, and there's no limit on how many times you use it, since scanning runs in your browser.
No. Scanning runs entirely in your browser. Nothing is transmitted, stored or logged.
No, and less so for Llama than for most models, since there's no single official interface it comes from. A finding tells you about the copy path, not the model.
Llama is open-weight and served through Meta's own apps plus dozens of unrelated third-party hosts, each with its own Markdown rendering and copy behavior. There is no one canonical Llama interface.
No. The scan looks at characters, not writing style, and makes no claim about which model produced the words.
No tool can promise that. Detectors judge sentence patterns and vocabulary, not invisible bytes.
No. This detector only scans and reports. Nothing is removed or rewritten here.
Yes, the site's watermark remover strips what this scan finds while leaving your wording untouched.
Yes. There's no length cap, since scanning runs on your own device rather than a server.
Yes. A zero-width space inside one splits the keyword, so the visible title and the slug or anchor stop matching.
Yes. Zero-width characters inside a snippet can stop it compiling with no visible cause, and the scan finds them.
The characters themselves are language-independent. The zero-width non-joiner is meaningful in Persian and some Indic scripts, so review those flags before removing anything.
Different hosting interfaces render and convert Markdown differently, which is the main reason the same underlying model can produce text with different hidden-character residue.
Not yet, on this tool. For several documents, paste, scan, copy the findings and repeat for now.
Yes. Nothing leaves your browser, so it's fine for drafts you haven't shared anywhere.
Yes. The page works the same in a mobile browser, no app required.
The scan itself is identical, the same characters checked the same way. This page is aimed at the specific search for Llama, since the many hosting interfaces are what shape what gets left behind, not the scan.
Paste, scan, read the findings. Nothing to install, nothing leaving your browser.
Scan my text