wiltodelta/remove-ai-watermarks
REMOVING AI WATERMARKS AND METADATA
Python library and CLI for removing visible and invisible AI watermarks, plus provenance metadata (C2PA, EXIF, XMP, IPTC, SynthID) from images, video, and audio. Supports Gemini, Sora, Veo, and other generator marks.
// KEY FEATURES
- Removes visible marks (Gemini, Sora, Veo) and invisible SynthID via diffusion
- Cleans C2PA, EXIF, XMP, IPTC metadata from photos, video, and audio
- CLI, Python API, and agent skill for Claude Code, Cursor, Codex
- Batch video processing and VAE regeneration for invisible watermarks
// DEEP DIVE: HOW IT WORKS
The tool provides a set of CLI commands (identify, visible, invisible, metadata, all, batch) and a Python API for detecting and removing AI provenance signals. Visible marks are removed via pixel manipulation, invisible SynthID via regeneration through calibrated VAE profiles, metadata via full stripping of C2PA/EXIF/XMP/IPTC manifests. Batch directory processing and video commands (video identify, video all, video invisible, video batch) are supported.
Inside: source classification across 10+ providers, detection of C2PA text/HTML wrappers, VSRMark, and Unicode steganography, automatic routing of Microsoft InvisMark to pixel regeneration, optional torch-free DWT-DCT detection. Video SynthID removal uses oracle-certified VAE regeneration; agent skill integrates into Claude Code, Codex, Cursor via npx skills.
Requires CUDA GPU for invisible watermarks and video SynthID (extra diffusion). Jurisdictions like China restrict AI label removal. Only for content you own; not for stock watermarks. Hosted raiw.cc offers free tier up to 12 MP for visible marks and metadata; original resolution and invisible removal are paid.
// USE CASES
Best for: Researchers and developers working with AI provenance and generative content
- Remove visible Gemini badge and C2PA metadata from generated image before publishing
- Clean batch of videos from invisible SynthID watermarks via VAE regeneration
- Find and erase text steganographic marks (VSRMark, Unicode) in LLM output
// QUICK START
uv tool install "remove-ai-watermarks[visible]"
npx skills add wiltodelta/remove-ai-watermarks
remove-ai-watermarks identify image.png
remove-ai-watermarks visible image.png
remove-ai-watermarks metadata image.pngCommands are taken from the repository README — check the current version on GitHub before running.
New repositories every 30 minutes
REDDYX AI scans GitHub 24/7 and ships the best AI/ML/Web3 projects to Telegram.
Join on Telegram