wiltodelta/remove-ai-watermarks

Remove AI Watermarks is a Python library and command-line tool for stripping visible and invisible AI watermarks and provenance metadata from images and video.

The tool addresses the problem of removing AI provenance marks from content you have generated yourself. It handles three categories of marks: visible labels like the Google Gemini sparkle watermark and vendor text, invisible pixel watermarks embedded through diffusion techniques, and structured metadata formats including C2PA, EXIF, XMP, and IPTC. For images, it offers commands to identify provenance signals, remove known visible marks, erase selected regions, strip metadata, or regenerate pixels to disrupt invisible watermarks. For video, it supports provenance identification, removal of visible marks from platforms like Sora and Kling AI, metadata stripping, and oracle-certified VAE regeneration for SynthID removal, with batch processing across directories.

Adoption depends on your specific needs and constraints. The tool suits developers and content creators who own generated media and want to remove provenance marks locally. Installation is modular: a minimal metadata-focused install requires no GPU, while invisible watermark removal requires CUDA. The project explicitly targets only content you own and does not target stock agency previews or third-party watermarks. A hosted service at raiw.cc offers the same functionality without local installation, with free tier limits on resolution and feature access.

Development shows consistent maintenance with regular updates addressing new watermark formats and video platform support. The project maintains detailed documentation including installation guides for different feature combinations, legal and safety notes clarifying scope, and guides for specific capabilities like photo classification. The codebase demonstrates active refinement of detection and removal techniques, with support for multiple diffusion models and video codecs expanding over time.