Watermark Tidy
Runs locally

Free · Private · No signup

Check and remove supported hidden watermarks

Inspect text for invisible Unicode and PNG/JPEG images for C2PA or supported AI metadata. Everything runs locally in your browser.

Not covered: statistical text, visible, or pixel watermarks.

Input

Check text or image metadata

Paste text to inspect invisible characters and unusual spacing.

Nothing is uploaded or stored.

Supported checks

What this watermark checker can find

A clean result means no supported signal was found. It does not prove that content has no watermark or that it was created by a person.

01

Hidden Unicode

Finds zero-width characters, direction controls, tags, variation selectors, and unusual spaces.

02

C2PA and AI metadata

Checks PNG and JPEG containers for Content Credentials and recognized AI-source metadata fields.

03

Important limits

Does not detect statistical text signals, visible overlays, SynthID pixels, or C2PA soft bindings.

How it works

Check first, remove only what you reviewed

The tool separates detection from removal. Your original text or image stays unchanged while you review supported findings and decide whether to create a cleaned copy.

  1. 1

    Choose text or an image

    Paste up to 500,000 characters or choose a PNG or JPEG up to 25 MB.

  2. 2

    Review the findings

    See matching character types, code points, positions, or image metadata containers.

  3. 3

    Download a cleaned copy

    Remove supported findings, re-check the result, then copy or download it.

Read the detection and removal method

Plain-language definition

What is a hidden watermark?

A hidden watermark is a signal that is not meant to be obvious in normal viewing. In text, that may mean invisible Unicode characters or statistical word-selection patterns. In images, it may mean metadata such as C2PA Content Credentials or a signal embedded in pixels. These methods require different detectors; no single check covers all of them.

Questions

Watermark checker and remover FAQ

Are hidden Unicode characters proof that text was written by AI?

No. Invisible characters can be legitimate in emoji, right-to-left writing systems, typography, or document formatting. A match shows which supported character exists; it does not identify who or what added it.

Does this checker detect SynthID or statistical text watermarks?

No. Statistical watermarks change token-selection patterns and require a compatible detector. This browser tool checks supported Unicode code points, not probability patterns in word choice.

Can this remove Claude's embedded text watermark?

No. It can remove supported invisible Unicode from text and supported C2PA structures from PNG or JPEG files. Anthropic describes its embedded text watermark as a separate signal, and its public documentation does not yet provide technical detector details.

What image watermark metadata can it check?

It checks supported PNG and JPEG structures for C2PA manifest stores, Content Credentials references, and recognized AI-source metadata fields. It does not perform full cryptographic C2PA signature or certificate validation.

Does removing supported image metadata change the pixels?

No. For supported removals, PNG image data and the JPEG compressed scan are copied byte-for-byte. The tool removes only identified metadata containers and re-checks the output before download.

Are my text and images uploaded?

No. Checks and removals run in your browser. The application has no upload endpoint, account system, analytics, or database. Loading the website itself still makes a normal request to its Cloudflare host.

Does a clean result prove an image or text has no watermark?

No. A clean result means only that no supported signal was found. Visible overlays, pixel-level marks, C2PA soft bindings, statistical text watermarks, and unknown vendor formats remain outside this check.

Technical terms follow the C2PA Content Credentials specification and the Unicode Standard. Last reviewed August 12, 2026.