A developer released watermarks-remover, an open-source, MIT-licensed tool that strips AI content markers, invisible Unicode text watermarks and file metadata from images, documents, audio and video, according to its GitHub repository.
The tool covers watermarking schemes from Claude, Gemini, OpenAI and open-source models, including Google's SynthID-Text, and ships as a Claude Code or Cursor agent skill, a dependency-free HTTP service built on Python's standard library, command-line scripts and a Docker deployment, the repository says.
The project's own documentation is blunt about the limits. Removing a statistical text watermark degrades the writing itself, because the watermark signal is spread across word choices throughout the text rather than added in one place. The developer writes that "no tool can honestly certify" a vendor's detector will fail, since detection APIs and keys are not public.
AI labs have leaned on invisible watermarks to prove content came from their models, a claim regulators and platforms increasingly want backed up. A tool built to strip that signal back out, positioned as a privacy feature for content you already own, is a preview of how much weight that proof can actually bear once removal is a free download.