pagedMark Tool Removes AI Provenance Watermarks from Images and Video
d0ofz · reddit · 2026-08-19
A developer released pagedMark, an open-source tool to strip AI provenance from generated content. It handles metadata like C2PA and EXIF, and removes invisible watermarks (e.g., SynthID) by regenerating pixels, which screenshots fail to do. It supports marks from ChatGPT, Sora, Kling, and others. Tests show processed GPT-Image images report 0 AI detection on OpenAI's verifier. The tool includes memory-aware processing for Apple Silicon to prevent swapping.
Related event: pagedMark Removes Invisible AI Watermarks from Images and Videos(2 posts)→
More from Safety
- Musk retweets discussion on AI oligopoly alignment — zacharylipton · 2026-08-19
- Zvi on Watermarks and Constraints: Not Primarily x-risk Reduction — TheZvi · 2026-08-19
- PA Governor Enacts Nation's Strictest AI Data Center Standards via Executive Order — TinfoilTricorn · 2026-08-19
- Stanford HAI hosts three-day AI boot camp for congressional staffers — StanfordHAI · 2026-08-19
- US Data Center Backlash: Public Ignores Medical and Defense Benefits of AI — Afinetheorem · 2026-08-19
- Critique on using safety as an excuse to farm user CoT and RL data — beffjezos · 2026-08-19