py-gemini-watermark-remover 0.1.0

Python implementation of a Gemini watermark removal tool using mathematical reverse alpha blending

Gemini Watermark Remover - Python Edition

A Python implementation of Gemini watermark removal tool using mathematical reverse alpha blending.

This project is a Python port of GeminiWatermarkTool.

中文文档

Demo

Original (Watermarked)	Cleaned

Features

• 🚀 Easy to use: Pure Python implementation, no compilation needed
• 🎯 Precise algorithm: Uses reverse alpha blending mathematics
• 📦 Minimal dependencies: Only requires OpenCV and NumPy
• 🔄 Batch processing: Supports single file and directory processing
• 🎨 Auto detection: Automatically detects watermark size (48x48 or 96x96)

Installation

Using uv (Recommended)

uv is a fast Python package and project manager.

# Install uv (if not already installed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install dependencies (creates virtual environment automatically)
uv sync

# Run directly
uv run python -m gemini_watermark_remover.cli image.jpg

Quick Start

Example Test

# Process example images
uv run python -m gemini_watermark_remover.cli -i examples/example1.jpg -o examples/example1_cleaned.jpg
uv run python -m gemini_watermark_remover.cli -i examples/example2.jpg -o examples/example2_cleaned.jpg

CLI Usage

Using uv (recommended):

# Simple mode - in-place edit (overwrites original!)
uv run python -m gemini_watermark_remover.cli watermarked.jpg

# Specify output file
uv run python -m gemini_watermark_remover.cli -i watermarked.jpg -o clean.jpg

# Batch process directory
uv run python -m gemini_watermark_remover.cli -i ./input_folder/ -o ./output_folder/

# Force watermark size
uv run python -m gemini_watermark_remover.cli -i image.jpg -o clean.jpg --force-small

# Show banner
uv run python -m gemini_watermark_remover.cli -i image.jpg -o clean.jpg --banner

Or using traditional method:

# After activating virtual environment
python -m gemini_watermark_remover.cli watermarked.jpg
python -m gemini_watermark_remover.cli -i watermarked.jpg -o clean.jpg

Python API

from gemini_watermark_remover import WatermarkRemover, process_image, process_directory
import cv2

# Method 1: Use convenience function for single file
process_image('watermarked.jpg', 'clean.jpg')

# Method 2: Use convenience function for directory
success, failed = process_directory('./input/', './output/')

# Method 3: Use WatermarkRemover class (more control)
remover = WatermarkRemover(logo_value=235.0)

# Read image
image = cv2.imread('watermarked.jpg')

# Remove watermark
cleaned = remover.remove_watermark(image)

# Save result
cv2.imwrite('clean.jpg', cleaned)

# Can also add watermark (for testing)
watermarked = remover.add_watermark(image)

Advanced Usage

from gemini_watermark_remover import WatermarkRemover, WatermarkSize
import cv2

# Create custom watermark remover
remover = WatermarkRemover(logo_value=235.0)

# Read image
image = cv2.imread('image.jpg')

# Force small watermark size
cleaned = remover.remove_watermark(
    image,
    force_size=WatermarkSize.SMALL
)

# Use custom alpha map
import numpy as np
custom_alpha = np.ones((48, 48), dtype=np.float32) * 0.5
cleaned = remover.remove_watermark(
    image,
    force_size=WatermarkSize.SMALL,
    alpha_map=custom_alpha
)

# Save
cv2.imwrite('output.jpg', cleaned, [cv2.IMWRITE_JPEG_QUALITY, 100])

CLI Arguments

Argument	Description
<file>	Simple mode: edit image in-place
-i, --input	Input file or directory
-o, --output	Output file or directory
-r, --remove	Remove watermark (default behavior)
--add	Add watermark (for testing)
--force-small	Force 48×48 watermark
--force-large	Force 96×96 watermark
--logo-value	Logo brightness value (default: 235.0)
-v, --verbose	Enable verbose output
-q, --quiet	Quiet mode
-b, --banner	Show ASCII banner
-V, --version	Show version
-h, --help	Show help

How It Works

Gemini Watermark Mechanism

Gemini adds watermarks using alpha blending:

watermarked = α × logo + (1 - α) × original

Reverse Alpha Blending Algorithm

Recover original pixels through mathematical inversion:

original = (watermarked - α × logo) / (1 - α)

Automatic Size Detection

Image Size	Watermark Size	Margin
W ≤ 1024 or H ≤ 1024	48×48	32px
W > 1024 and H > 1024	96×96	64px

API Reference

WatermarkRemover Class

class WatermarkRemover:
    def __init__(self, logo_value: float = 235.0)

    def remove_watermark(
        self,
        image: np.ndarray,
        force_size: Optional[WatermarkSize] = None,
        alpha_map: Optional[np.ndarray] = None
    ) -> np.ndarray

    def add_watermark(
        self,
        image: np.ndarray,
        force_size: Optional[WatermarkSize] = None,
        alpha_map: Optional[np.ndarray] = None
    ) -> np.ndarray

    @staticmethod
    def get_watermark_size(width: int, height: int) -> WatermarkSize

    @staticmethod
    def calculate_alpha_map(bg_capture: np.ndarray) -> np.ndarray

Convenience Functions

def process_image(
    input_path: Union[str, Path],
    output_path: Union[str, Path],
    remove: bool = True,
    force_size: Optional[WatermarkSize] = None,
    logo_value: float = 235.0
) -> bool

def process_directory(
    input_dir: Union[str, Path],
    output_dir: Union[str, Path],
    remove: bool = True,
    force_size: Optional[WatermarkSize] = None,
    logo_value: float = 235.0
) -> Tuple[int, int]

Supported Image Formats

• JPEG (.jpg, .jpeg)
• PNG (.png)
• WebP (.webp)
• BMP (.bmp)

Project Structure

py-gemini-watermark-remover/
├── assets/
│   ├── bg_48.png
│   └── bg_96.png
├── src/
│   └── gemini_watermark_remover/
│       ├── __init__.py
│       ├── cli.py
│       └── watermark_remover.py
├── tests/
│   └── test.py
├── examples/
│   ├── example1.jpg
│   ├── example1_cleaned.jpg
│   ├── example2.jpg
│   └── example2_cleaned.jpg
├── README.md
├── README_zh.md
└── pyproject.toml

Performance

• Single image processing: ~200-800ms (depends on image size and hardware)
• Batch processing: Sequential processing of multiple files
• Memory usage: ~3-4x image size (for floating-point operations)

Limitations

• Only removes visible watermarks (bottom-right semi-transparent logo)
• Does not remove hidden/steganographic watermarks
• Designed for Gemini's current watermark pattern (2025)

Troubleshooting

Issue: Processed image looks unchanged

The watermark is semi-transparent. If the background color is similar to the watermark, the difference may be subtle. Zoom to 100% and check the bottom-right corner.

Issue: Wrong watermark size detected

Use --force-small or --force-large to manually specify:

uv run python -m gemini_watermark_remover.cli -i image.jpg -o clean.jpg --force-small

Issue: ModuleNotFoundError

Make sure dependencies are installed:

uv sync

Comparison with C++ Version

Feature	C++ Version	Python Version
Installation	No installation (single file)	Requires Python environment
File size	~15MB	~2KB (excluding dependencies)
Speed	Fast	Medium (NumPy optimized)
Code size	~1000 lines	~600 lines
Development	Requires compilation	Edit and run
Easy to modify	Medium	Easy
Best for	Distribution to users	Development/Integration

License

MIT License

Disclaimer

This tool is for personal and educational use only. Users must ensure their use complies with applicable laws and terms of service.

The author is not responsible for any data loss or image corruption caused by using this tool. Please backup original images before use.

Credits

Python implementation based on GeminiWatermarkTool C++ version.

---

If this tool helped you, please give the project a ⭐