A Python framework for building LLM-powered applications and agents, with standard interfaces for models, embeddings, vector stores and third-party integrations.
guillaumemeyer/watermarks-remover
Agent skill and Python service that strip AI provenance marks like C2PA metadata and invisible Unicode from your files.
- Stars
- 24,411
- Forks
- 2,801
- Open issues
- 22
- Last commit
- 4 days ago
What is it?
watermarks-remover is an agent skill backed by a stdlib-only Python HTTP service. It inspects and cleans text and files of AI provenance marks: invisible Unicode characters in text, and C2PA manifests plus EXIF, XMP or document metadata in images, PDFs, Office files, audio and video. The skill is a thin client, so the agent host does not need Python installed.
Text cleaning has two layers. Deterministic removal handles invisible characters and unusual spaces, and an optional rewrite step, run through a local Ollama model or an OpenAI-compatible API, targets statistical token-sampling watermarks. Detection of several text and image watermark schemes runs as separate steps that do not block cleaning. Optional SynthID pixel scoring and CtrlRegen pixel removal load from user-provided checkouts.
Key features
- 01Removes invisible Unicode, unusual spaces and bidi characters from text.
- 02Strips C2PA manifests and metadata from images, audio, video and documents.
- 03Exposes inspect, detect, clean and watermark endpoints over HTTP with an OpenAPI spec.
- 04Installs into Claude Code, Codex, Cursor and other agent hosts with one installer.
- 05Provides a PostToolUse hook that checks or cleans files the agent writes.
- 06Supports batch endpoints and published Docker images.
Use cases
- Removing provenance metadata from your own photos or documents before sharing them.
- Cleaning hidden characters from AI-assisted drafts of manuscripts or web copy.
- Adding a pre-commit or CI check that flags AI marks in repository files.
- Testing whether a text watermark survives a rewrite in controlled experiments.
Quick start
- 1
Clone the repository
- 2
Start the local HTTP service (Python 3.10+, no dependencies)
make serve - 3
Install the agent skill for Claude Code
python3 install_skill.py --skill remove-ai-marks --target claude-code - 4
Clean a file using the scripts
python3 service/scripts/clean_file.py draft.md -o draft.cleaned.md
Things to know
- The skill is a thin client and needs the Python service running at WATERMARKS_SERVICE_URL.
- Statistical text watermark removal (Layer B) relies on agent rewriting and is best-effort.
- Full PDF stripping requires the qpdf system tool, and c2patool and exiftool are optional extras.
Topics
FAQ
- What is watermarks-remover?
- Agent skill and Python service that strip AI provenance marks like C2PA metadata and invisible Unicode from your files. watermarks-remover is an agent skill backed by a stdlib-only Python HTTP service. It inspects and cleans text and files of AI provenance marks: invisible Unicode characters in text, and C2PA manifests plus EXIF, XMP or document metadata in images, PDFs, Office files, audio and video. The skill is a thin client, so the agent host does not need Python installed.
- Is watermarks-remover free to use? What license does it have?
- watermarks-remover is open source under the MIT license. Check the license terms before using it in commercial products.
- Is watermarks-remover still maintained?
- Actively maintained. The last commit to watermarks-remover was on Oct 17, 2026.
- What can I use watermarks-remover for?
- Removing provenance metadata from your own photos or documents before sharing them. Cleaning hidden characters from AI-assisted drafts of manuscripts or web copy. Adding a pre-commit or CI check that flags AI marks in repository files. Testing whether a text watermark survives a rewrite in controlled experiments.
- How do I install watermarks-remover?
- Clone the repository Start the local HTTP service (Python 3.10+, no dependencies): make serve Install the agent skill for Claude Code: python3 install_skill.py --skill remove-ai-marks --target claude-code Clean a file using the scripts: python3 service/scripts/clean_file.py draft.md -o draft.cleaned.md
- Where can I get watermarks-remover?
- The source code and installation instructions are on GitHub: https://github.com/guillaumemeyer/watermarks-remover
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
A command-line tool and Python library for running prompts and chats against remote and locally installed large language models.
A minimal Python library for building AI agents that write and run their actions as Python code.
TypeScript, Python and Go SDK for browser agents that drives Chrome with natural-language act, observe and extract calls alongside Playwright-style APIs.
Open-source browser automation that uses LLMs and computer vision to run web workflows on any site, extending Playwright with AI commands.
Agent-native memory infrastructure that turns LLM agent conversations and execution into persistent, structured memory for Python and TypeScript apps.