GitHub Tool Removes Invisible AI Watermarks Like SynthID and Metadata

tom_doerr · x · 2026-08-19

GitHub user wiltodelta released remove-ai-watermarks, an open-source project to strip AI provenance marks from images and videos. It supports removing visible labels (e.g., Gemini sparkle), invisible pixel watermarks via diffusion regeneration (e.g., SynthID), and metadata standards like C2PA, EXIF, XMP, and IPTC. The tool provides a Python library and CLI, featuring batch processing for videos and capabilities to clear visible marks from models like Sora, Veo, and Kling.

Original post →

More from Apps

Apps channel →