AI 驱动的图片去水印工具 | AI-Powered Image Watermark Remover
🚀 秒级处理 · 🔒 本地运行 · 💯 完全免费 · ✨ 高质量输出
Clean PicGo 是一款基于 AI 技术的在线图片去水印工具。使用先进的 LaMa (Large Mask Inpainting) 深度学习模型，可以智能地去除图片中的水印、文字和不需要的物体。
🌟 核心亮点：所有处理完全在浏览器本地进行，图片不会上传到任何服务器，充分保护您的隐私！
访问 clean.picgo.studio 立即体验！
| 功能 | 描述 |
|---|---|
| 🎨 智能擦除 | 使用橡皮擦工具选中需要去除的区域，AI 会自动填充背景 |
| ⚡ Gemini 专用模式 | 自动检测并无损去除 Gemini AI 生成图片的水印，基于反向 Alpha 混合算法 |
| 🧵 Web Worker 后台处理 | 模型推理在独立线程运行，主线程完全不阻塞，处理期间可自由操作 |
| 🔒 隐私安全 | 所有处理完全在本地进行，图片不会上传到任何服务器 |
| 🎭 蒙版技术 | 精确的蒙版生成技术，确保只处理选中的区域 |
| 👁️ 实时对比 | 处理完成后可以左右对比查看处理前后的效果差异 |
| 📥 高清导出 | 支持导出原始分辨率的处理结果，保持图片质量 |
| 📱 PWA 支持 | 可安装为桌面应用，支持离线使用 |
针对 Gemini AI 生成图片的水印，Clean PicGo 提供了专门的无损去除方案：
- 自动检测：上传图片后自动检测是否存在 Gemini 水印
- 无损去除：使用数学精确的反向 Alpha 混合算法，而非 AI 推理
- 毫秒级处理：无需加载大模型，处理速度极快
- 可回退操作：自动处理作为一个可回退的编辑步骤
此功能基于 gemini-watermark-remover 项目的算法原理实现。
- 上传图片 - 拖拽或点击上传需要处理的图片（支持 JPG、PNG、WebP）
- 自动检测 - 如果是 Gemini 图片，系统会自动检测并去除水印
- 选择区域 - 使用橡皮擦工具涂抹需要去除的其他水印区域
- AI 处理 - LaMa 模型自动分析并填充选中区域
- 下载结果 - 预览效果满意后下载处理后的图片
| 类别 | 技术 |
|---|---|
| 框架 | Next.js 16 + React 19 |
| 语言 | TypeScript 5 |
| 样式 | Tailwind CSS 4 |
| AI 推理 | ONNX Runtime Web + Web Worker |
| AI 模型 | LaMa (Large Mask Inpainting) |
| 动画 | Framer Motion |
| 国际化 | next-intl |
| UI 组件 | Radix UI |
# 克隆项目
git clone https://github.com/isboyjc/cleanup.git
cd cleanup
# 安装依赖
pnpm install
# 启动开发服务器
pnpm dev
# 构建生产版本
pnpm build
# 启动生产服务器
pnpm start访问 http://localhost:3000 查看应用。
Clean PicGo 支持以下语言：
- 🇨🇳 简体中文
- 🇺🇸 English
支持浅色模式和深色模式，可跟随系统设置自动切换。
Clean PicGo is an AI-powered online image watermark removal tool. Using the advanced LaMa (Large Mask Inpainting) deep learning model, it can intelligently remove watermarks, text, and unwanted objects from images.
🌟 Key Highlight: All processing is done entirely in your browser locally. Images are never uploaded to any server, fully protecting your privacy!
Visit clean.picgo.studio to try it now!
| Feature | Description |
|---|---|
| 🎨 Smart Erase | Select the area to remove with the eraser tool, and AI will automatically fill in the background |
| ⚡ Gemini Mode | Auto-detect and losslessly remove watermarks from Gemini AI images using Reverse Alpha Blending |
| 🧵 Web Worker Processing | Model inference runs in a separate thread, keeping the main thread responsive during processing |
| 🔒 Privacy Safe | All processing is done locally, images are never uploaded to any server |
| 🎭 Mask Technology | Precise mask generation ensures only selected areas are processed |
| 👁️ Real-time Comparison | Compare before and after effects with a side-by-side slider |
| 📥 HD Export | Export results at original resolution, maintaining image quality |
| 📱 PWA Support | Install as a desktop app with offline support |
Clean PicGo provides a dedicated lossless removal solution for Gemini AI-generated image watermarks:
- Auto Detection: Automatically detects Gemini watermarks after image upload
- Lossless Removal: Uses mathematically precise Reverse Alpha Blending algorithm instead of AI inference
- Instant Processing: No large model loading required, extremely fast processing
- Undoable Operation: Auto-processing is saved as an undoable editing step
This feature is based on the algorithm from gemini-watermark-remover project.
- Upload Image - Drag and drop or click to upload an image (supports JPG, PNG, WebP)
- Auto-Detection - If it's a Gemini image, system auto-detects and removes watermark
- Select Area - Use the eraser tool to paint over other watermark areas
- AI Processing - LaMa model automatically analyzes and fills the selected area
- Download Result - Preview the result and download the processed image
| Category | Technology |
|---|---|
| Framework | Next.js 16 + React 19 |
| Language | TypeScript 5 |
| Styling | Tailwind CSS 4 |
| AI Inference | ONNX Runtime Web + Web Worker |
| AI Model | LaMa (Large Mask Inpainting) |
| Animation | Framer Motion |
| i18n | next-intl |
| UI Components | Radix UI |
# Clone the repository
git clone https://github.com/isboyjc/cleanup.git
cd cleanup
# Install dependencies
pnpm install
# Start development server
pnpm dev
# Build for production
pnpm build
# Start production server
pnpm startOpen http://localhost:3000 to view the application.
Clean PicGo supports the following languages:
- 🇨🇳 Simplified Chinese
- 🇺🇸 English
Supports light and dark mode, can automatically switch based on system settings.
