New Open-Source Tool Strips AI Watermarks From Text and Images, Sparks Developer Debate Over Ethics and Limitations

A new open-source tool called 'watermarks-remover' is surging on GitHub with 4.8k stars, offering developers the ability to strip AI watermarks from text and images across multiple file formats — but creators warn it's best-effort only, lossy on statistical watermarks, and intended for legitimate privacy use, not fraudulent claims of human-authored content.

• A new open-source GitHub project called 'watermarks-remover' is gaining traction with 4.8k stars, offering Python scripts and an agent skill to strip multi-vendor AI provenance marks — including invisible Unicode characters, statistical token-sampling watermarks, and C2PA/EXIF/XMP metadata — from file types such as PNG, JPEG, SVG, PDF, DOCX, and Markdown.
• The tool operates across three layers: Layer A deterministically removes edit-based Unicode carriers, Layer B performs best-effort statistical watermark removal via text rewriting, and a file layer strips C2PA and metadata from supported containers, with optional external backends for pixel-domain image watermark removal using CtrlRegen and SynthID scoring.
• Developers are cautioning that Layer B text watermark removal is inherently lossy and best-effort, as stripping statistical marks requires substantial rewording that degrades tone and quality, and no tool can certify that official vendor detectors will fail — making the project best suited for privacy and hygiene on content users own, not for fraudulent 'human-written' claims.