Claude Watermark Remover & Checker for Text and Files
Detect hidden Unicode characters in Claude text and inspect PNG, JPG, WebP, or SVG files for supported C2PA and AI provenance metadata. Remove supported marks locally in your browser, verify the cleaned result, and download a separate copy without changing the original.
How to use this tool
- 1
Choose Text or File
Open the text checker for pasted writing, or use the file checker for PNG, JPG, WebP, and SVG output. Each mode uses a separate inspection path.
- 2
Run the Watermark Check
The text mode counts supported invisible characters without deleting language controls automatically. File mode looks for C2PA, JUMBF, Anthropic, Claude, and AI provenance markers inside supported containers.
- 3
Review the Findings
Read what was detected, how many items were found, and whether each item can be removed conservatively or must be preserved for multilingual safety.
- 4
Remove Supported Marks
Clean supported hidden text characters or remove identified provenance chunks and metadata. Images are not uploaded or visually regenerated.
- 5
Verify and Download
The cleaned output is checked again. Copy or download cleaned text, or save a separately named cleaned file while keeping the original untouched.
What it can do
Checker and Remover Together
Inspect first, remove only supported findings, and see a post-clean verification result in one workspace instead of trusting a blind conversion.
Multilingual-Safe Text Cleaning
The conservative cleaner removes a small set of known hidden characters while preserving direction controls, tag sequences, joiners, and other characters that may be valid in Hindi, Arabic, emoji, or complex scripts.
C2PA and JUMBF Inspection
The file checker reads PNG chunks, JPEG application segments, WebP RIFF chunks, and SVG metadata to locate supported provenance structures.
No Image Re-Encoding
Metadata cleaning keeps the original compressed image data. The tool does not redraw, resize, recompress, or deliberately change visible pixels.
Honest Statistical-Watermark Status
The result distinguishes removable Unicode or metadata from Claude's model-level statistical text mark, which cannot yet be independently verified with an official public detector.
Before-and-After Evidence
Finding counts and file sizes are shown before and after cleaning so you can see what the tool changed and what it deliberately left alone.
Private Browser Workflow
Text and files stay on the device for inspection, cleaning, verification, copying, and download. No account or upload is required.
Useful for
- Checking Claude-assisted writing for hidden copy-and-paste characters
- Removing supported C2PA provenance from an authorized Claude-generated image
- Cleaning AI metadata from SVG assets before adding them to a codebase
- Verifying whether a downloaded PNG or JPEG contains supported provenance chunks
- Preparing owned files for a privacy-sensitive internal workflow
- Comparing the original and cleaned file size before download
Limits to know
- Anthropic has not released its official statistical text-watermark detector or detection keys
- No result can guarantee that every private Claude or third-party detector will fail
- Heavy paraphrasing is not included because it can change tone, facts, citations, code, and SEO copy
- The first browser version supports text, PNG, JPG, WebP, and SVG; PDF, DOCX, audio, and video are not accepted
- Pixel-domain watermarks, soft-bound provenance, screenshots, and remote manifests may remain after metadata cleaning
- Removing provenance does not prove human authorship and should not be used for academic fraud or false disclosure
Private browser processing
This tool runs locally in your browser. TunerPage does not upload the working file or text to its servers.
Frequently asked questions
What is the Claude Watermark Remover & Checker?
It is a combined browser tool that checks text for supported hidden Unicode and checks PNG, JPG, WebP, or SVG files for C2PA and AI provenance metadata. It can remove supported findings and recheck the result.
Does Claude really add watermarks to text?
Anthropic says supported Claude models use an imperceptible model-level watermark embedded in generated text. This is a statistical token pattern, not simply a zero-width character or file property.
Can this tool detect Claude's statistical text watermark?
Not conclusively. Anthropic has not yet published the official detector, algorithm details, or keys. The text checker can find supported hidden Unicode, but it labels the model-level statistical signal as unverifiable.
Can it remove C2PA metadata from Claude images?
It can identify and remove supported C2PA, JUMBF, and AI provenance chunks from PNG, JPG, WebP, and SVG files. The cleaned result is inspected again before download.
Will image quality change?
The metadata cleaner preserves the existing compressed image payload and does not re-encode visible pixels. Removing metadata can change the file bytes and invalidate a provenance signature, but it should not visually recompress the image.
Why are some hidden text characters preserved?
Direction marks, tag characters, and related controls can be legitimate parts of multilingual writing, emoji, or complex scripts. The checker reports them but the conservative cleaner leaves them in place to avoid damaging valid text.
Are my text and files uploaded?
No. Supported inspection, cleaning, verification, copying, and downloads happen locally in the browser. The working content is not sent to TunerPage.
Does a clean result prove the content is human-written?
No. Missing metadata or removable characters says nothing conclusive about authorship. Human writing can be processed by Claude, and AI-generated content can lack detectable marks.
Which files are supported?
This version supports PNG, JPG, JPEG, WebP, and SVG files up to 25 MB. PDF, DOCX, ODT, audio, and video require different structural tools and are not accepted in this release.