gemini-unwatermark 0.1.0

Remove watermarks from Gemini-generated images using LaMa inpainting

gemini-unwatermark

Fast CLI to remove watermarks from Google Gemini-generated images using AI inpainting.

Installation

pip install gemini-unwatermark

Or run directly without installing:

uvx gemini-unwatermark image.png

For GPU acceleration (CUDA):

pip install "gemini-unwatermark[gpu]"

Requirements: Python 3.10-3.13

Usage

# Single image
gemini-unwatermark image.png

# Custom output
gemini-unwatermark image.png -o clean.png

# Batch processing
gemini-unwatermark *.png -d ./cleaned/

# Parallel (8 workers)
gemini-unwatermark photos/*.jpg -d ./output -w 8

Options

Option	Description
-o, --output	Output path (single image only)
-d, --output-dir	Output directory for batch
-f, --force	Overwrite existing files
-q, --quiet	Suppress progress output
-w, --workers	Parallel workers (1-8, default: 4)
--model-path	Custom ONNX model path

How it works

Uses LaMa (Large Mask Inpainting) to intelligently fill the watermark region in the bottom-right corner of Gemini images.

The model (~200MB) auto-downloads on first use to ~/.cache/gemini-unwatermark/.

Disclaimer

This tool only removes the visible watermark. It does not remove SynthID, Google's invisible watermarking technology embedded in the image data.

Credits

• Original Chrome extension: dinoBOLT/Gemini-Watermark-Remover
• LaMa ONNX model: Carve/LaMa-ONNX

License

Apache-2.0