Remove-AI-Watermarks - A CLI and library for removing AI watermarks from images

Remove-AI-Watermarks is a CLI and Python library that handles visible watermarks, invisible watermarks, and metadata in AI-generated images.

It targets services such as Google Gemini/Nano Banana, ChatGPT/DALL-E, Stable Diffusion, Adobe Firefly, and Midjourney, aiming to remove each service's watermarks, signatures, and EXIF/XMP information.

• Visible watermark removal: Targets the sparkle logo from Gemini/Nano Banana, using alpha blending inversion and inpainting.
• Invisible watermark removal: Weakens patterns like SynthID, StableSignature, and TreeRing through diffusion-based regeneration.
• Metadata cleanup: Handles EXIF, PNG text chunks, XMP DigitalSourceType, and C2PA Content Credentials, removing fields that trigger platforms' "Made with AI" labels.
• Face protection feature: Provides Smart Face Protection, which detects people with YOLO and then re-blends the original face to reduce distortion.
• Additional options: Also includes Analog Humanizer, which adds film grain and chromatic aberration.

Installation is available via pipx or uv tool, with the default install including visible watermark removal and metadata removal. Invisible watermark removal requires Python 3.10+, and downloads a roughly 2GB model on first run.

The roadmap includes SynthID-Image v2 regression testing, improving limitations in internal metadata removal for AVIF/HEIF/JPEG-XL, and plans to separate out a video pipeline. However, it states that it does not support removal of Nightshade, Glaze, or PhotoGuard, viewing that as falling within the scope of attacking artist protections.