Remove AI watermarks from content you own
watermarks-remover strips multi-vendor AI provenance marks - invisible Unicode, statistical text watermarks, and C2PA/EXIF/XMP metadata - from your text and files. 100% local, open source, no uploads.
Privacy-first - everything runs on your machine
What is watermarks-remover?
An open-source agent skill and stdlib-only Python service that removes AI provenance marks from text and files - for privacy and hygiene on content you own.
Layer A - Text hygiene
Deterministic scripts strip invisible Unicode, exotic spaces, bidi controls and tag characters from text.
Layer B - Statistical rewrites
Rewrite hooks neutralise token-sampling (Kirchenbauer-style) text watermarks used by major AI vendors.
File metadata cleanup
Clears C2PA / EXIF / XMP and document properties from PNG, JPEG, WebP, SVG, PDF, DOCX, HTML, Markdown and more.
Multi-vendor support
Targets Claude, Gemini / SynthID, OpenAI and open-LLM provenance surfaces in one tool.
Why watermarks-remover
Privacy, coverage and control - without the complexity.
How it works
Inspect, then clean - everything stays on your machine.
Install
Install the skill and start the service - Python 3.10+ stdlib only, no dependencies, no Docker required.
Inspect
Run inspect_file.py or inspect_text.py to see exactly which AI marks and metadata are present.
Clean
Run clean_file.py or clean_text.py to strip the marks and save a clean copy.
Verify
Re-inspect the output to confirm the provenance marks are gone.
Key features
Everything you need to strip AI provenance marks from content you own.
Invisible Unicode cleanup
Removes zero-width characters, exotic spaces, bidi and tag characters that fingerprint AI text.
Statistical watermark neutralisation
Rewrite hooks for token-sampling text watermarks from Claude, Gemini / SynthID and OpenAI.
C2PA / EXIF / XMP stripping
Clears provenance and metadata from PNG, JPEG, WebP, SVG, PDF, DOCX and more.
Inspect before you clean
See exactly which marks are present before touching your files.
Optional deep-strip tools
Auto-uses c2patool, exiftool and qpdf when present for residual metadata and PDF rebuilds.
Open source & auditable
MIT licensed and transparent, with docs covering residual risk and verification.
Frequently asked questions
Have another question? Open an issue on GitHub.
Take control of your content
Star the project and start stripping AI provenance marks locally today.