gemini_watermark_cleaner/README.md

[中文文档](README_zh.md)

# Gemini & NotebookLM Watermark Cleaner

A privacy-first, browser-based tool for removing watermarks from Gemini and NotebookLM generated content (Images & PDFs). Built with vanilla JavaScript, featuring mathematical precision for lossless restoration and robust fallback cloning for complex document formats.

<p align="center">
  <img src="https://count.getloli.com/@gemini-watermark-remover?name=gemini-watermark-remover&theme=minecraft&padding=7&offset=0&align=top&scale=1&pixelated=1&darkmode=auto" width="400">
</p>

## ✨ Features

- **Multi-Format Support**: Handles standard Gemini Images (`.jpg`, `.png`, `.webp`) and NotebookLM documents (`.pdf`).
- **Lossless Restoration**: Uses **Reverse Alpha Blending** for standard Gemini images to perfectly restore original pixels without blurring.
- **Smart Cloning**: Applies intelligent **Simple Clone** (patching) for NotebookLM headers/footers and PDF slides where alpha maps don't apply.
- **Privacy-First**: 100% client-side processing. No files are ever uploaded to any server.
- **Batch Processing**: Process multiple files at once and download them as a ZIP archive.
- **Modern UI**: Dark/Light mode support with a clean, responsive interface.
- **PDF Preview**: Real-time progress tracking and preview for PDF file cleaning.

## 🛠️ Technology Stack

- **Core Logic**: Vanilla JavaScript (ES Modules)
- **PDF Processing**: `pdfjs-dist` (Parsing) & `jspdf` (Re-generation)
- **UI Styling**: Tailwind CSS (CDN)
- **Packaging**: `esbuild` for bundling

## 🧩 How It Works

This tool employs two distinct strategies depending on the content type:

### 1. Reverse Alpha Blending (for Gemini Images)
Gemini applies watermarks using a standard alpha compositing formula:
$$Pixel_{watermarked} = \alpha \cdot Pixel_{logo} + (1 - \alpha) \cdot Pixel_{original}$$

We reverse this to recover the original pixel:
$$Pixel_{original} = \frac{Pixel_{watermarked} - \alpha \cdot Pixel_{logo}}{1 - \alpha}$$

By using pre-calibrated Alpha Maps (48px/96px), we can mathematically restore the original pixel values with near-zero loss.

### 2. Smart Cloning (for NotebookLM & PDF)
For NotebookLM's strip-like watermarks or PDF slides where exact alpha maps vary, the tool uses a **Simple Clone** technique. It identifies the watermark region and intelligently "clones" a patch of pixels from the immediate neighborhood (e.g., just above the watermark) to cover it, blending the edges for a seamless look.

## 🔗 Attribution & References

This project is built upon the excellent work of the open-source community. Special thanks to:

- **Original Project**: [gemini-watermark-remover](https://github.com/journey-ad/gemini-watermark-remover) by [journey-ad](https://github.com/journey-ad).
  - *Base architecture, alpha map logic, and core "Reverse Alpha Blending" implementation.*

- **Algorithm Origin**: [Gemini Watermark Tool](https://github.com/allenk/GeminiWatermarkTool) by [Allen Kuo](https://github.com/allenk).
  - *Original discovery of the math and watermark masks.*

## 🚀 Usage

### Development
1. Clone the repository:
   ```bash
   git clone http://git.unissense.tech/mula/gemini_watermark_cleaner.git
   ```
2. Install dependencies:
   ```bash
   pnpm install
   ```
3. Start development mode (rebuilds on changes):
   ```bash
   pnpm dev
   ```
4. Start a local preview server:
   ```bash
   pnpm serve
   ```
5. Build for production:
   ```bash
   pnpm build
   ```

### Docker Deployment
1. Build and run the container:
   ```bash
   docker-compose up -d --build
   ```
2. The application will be available at `http://localhost:3000` (or the port specified in `.env`).

### Browser
Simply open the `dist/index.html` (after building) or the deployed site. Drag and drop your files.
- **Images**: Instant preview and cleaning.
- **PDFs**: Shows progress per page, then displays the cleaned first page.

## 📄 License

[MIT License](./LICENSE)