Tools
A 931-star repo for stripping AI provenance marks lands three days after launch, targeting Unicode, statistical watermarks and C2PA
Leutenegger/watermarks-remover appeared 2026-08-19 and is at 931 stars. It is structured as an agent skill plus a stdlib Python HTTP service across three layers: deterministic scripts for invisible Unicode, exotic spaces, bidi and tag characters; an agent rewrite pass aimed at token-sampling statistical watermarks; and metadata and C2PA stripping across PNG, JPEG, SVG, PDF, DOCX, HTML and Markdown. The repo frames this as privacy hygiene for content you own. Star velocity at this level is the signal worth reading: content-provenance schemes now have an organized open-source counterparty, which is a real input to anyone betting on C2PA as a detection layer.
Source
↳ Follow the thread