A CLI and Python library that removes visible and invisible AI watermarks, provenance metadata, and arbitrary regions from AI-generated images using inpainting. It supports batch processing and is installable locally, though some removal tasks require downloading external model weights.
Project overview
It combines visible watermark removal, invisible watermark stripping, and metadata erasure into a single command-line and Python interface for AI-generated images.
Project type
Image & Vision
Use cases
Image Processing
Deployment
Refer to project documentation
License
Apache-2.0
Best for
Developers needing to strip AI watermarks and provenance metadata from AI-generated images via a CLI or library.
Key capabilities
Detects and removes known visible AI watermarks such as the Gemini sparkle and Doubao text strip using localized inpainting via cv2, MI-GAN, or big-LaMa.
Provides a generic inpainting tool to erase arbitrary logos or objects from a specified bounding box.
Processes entire directories of images in a single run.
Limitations and risks
Samsung Galaxy AI detection is locale-specific and currently detects only the Italian variant.
Invisible watermark removal requires downloading SDXL models, which are approximately 2 GB in size.
Getting started
Core CLI features are installable via pip or pipx. A documented first success path is installing via pipx and running the visible watermark removal command on a target image.
Evidence and sources
GitHub project description: AI watermark remover. CLI and Python library to strip visible and invisible AI watermarks (Gemini / Nano Banana sparkle, SynthID) and provenance metadata (C2PA, EXIF, IPTC) from i…
README: Remove **visible** and **invisible** AI watermarks from images generated by Google Gemini