wtf is watermarks-remover?

watermarks-remover is a toolkit for stripping AI-related provenance marks and hidden signals from files someone owns, built both as scripts a person can run directly and as a skill for AI coding assistants like Claude Code or Grok. It targets three separate kinds of markers that AI generated content can carry: invisible Unicode characters and unusual spacing that some tools quietly insert into text, statistical patterns hidden in word choice that services like Google's SynthID use to fingerprint generated text, and embedded metadata standards like C2PA that record where an image or document came from. The invisible Unicode and metadata removal is handled by plain Python scripts using only the standard library, with no external dependencies required for basic use, covering file types including PNG, JPEG, SVG, PDF, DOCX, and Markdown. Removing statistical text watermarks is described as much harder, since that kind of signal is spread across nearly every word choice rather than hidden in a few extra characters, so the tool's approach is to have an AI model rewrite the text, which the README is explicit will change the tone and quality of the writing, not act as a perfect eraser. The project can optionally connect to an external, separately licensed tool to estimate how likely an image is to carry Google's SynthID watermark, either by installing that scorer locally or running it inside a Docker container, though this scoring can only detect a likely watermark, not remove pixel-level marks from images. Installation involves linking the skill folder into an AI assistant's configuration directory, after which it can be invoked with a command like "/remove-ai-marks" or a plain request to strip watermarks from a file. The core scripts only need Python 3.10 or newer with no extra libraries. Optional metadata tools like exiftool improve results on PDFs specifically.