Scrub hidden AI watermarks and C2PA metadata from your files

A new open-source tool lets you scrub AI watermarks and hidden tracking signatures from your files.

Developer Guillaume Meyer released watermarks-remover, a project that accumulated over 13,000 GitHub stars this week. It cleans invisible Unicode characters, C2PA signatures, and metadata out of text, images, PDFs, and Office documents. The core tool relies strictly on Python's standard library, meaning it runs locally without extra dependencies.

Why it matters: AI models from OpenAI, Claude, and Google embed subtle watermarks and metadata into generated outputs. This tool gives you a fast, local way to sanitize content you own, wipe vendor tracking, and preserve digital privacy before publishing.

Here's the gist: run the scripts directly on files like PNGs, PDFs, or Markdown, or spin up its Python HTTP service on port 8765. It handles multi-vendor marking schemes—including Gemini's SynthID-Text and open-LLM token sampling—without needing heavy setup or external APIs.

Your files, cleaned down to the last hidden byte.