---

💡 Why Kites?

Traditional manga translators require installing complex Python scripts, gigabytes of CUDA packages, or uploading your private reading history to third-party cloud servers.

Kites changes that. It is a lightweight Chrome extension that runs state-of-the-art vision and language models directly on your graphics card via WebGPU.

• ⚡ Zero Setup: Install with 1-click from the Chrome Web Store and start reading immediately. No terminal, no Python, no dependencies.
• 🔒 100% Local & Private: Optical character recognition (OCR) and background cleaning happen on your device. Your images never leave your browser.
• 🎨 Intelligent Background Repair: AI inpainting erases foreign text cleanly while preserving linework, screentones, and textures.
• ✍️ Natural Typesetting: Automatically measures speech balloons, balances line breaks, fits font sizes, and adapts font colors.
• 🗨️ Smart Speech Bubble Fit: Advanced bubble detection (Heuristic & Neural YOLO) expands narrow CJK columns into roomy balloon chambers, eliminating tiny squeezed fonts.
• 🌈 Color-Bubble Detection & Wrap: Detects per-bubble text color automatically, matches font color, and wraps translated text inside original speech balloon shapes.
• 📜 Full Webtoon Strip Support: Seamlessly translates extreme long-strip webtoons ($>14,000\text{ px}$) with 1D in-memory canvas tiling at full native resolution.
• 🛠️ Built-in Studio: Fix mistranslations, resize text balloons, or tweak typography on the fly.

---

🖼️ See It in Action

Full Translation + Colorization (Chinese → English)

Original grayscale Chinese page alongside the fully translated and colorized English result, with speech balloon wrapping. Artwork: @almondnuts117

Japanese → English

Story: 佐藤さんは知っていた by @09ra_19ra

Japanese → Vietnamese

Source: Pixiv Artwork #145272482

---

🖱️ Instant In-Place Translation

Hover over any comic panel on Twitter/X, Pixiv, or manga sites. An unobtrusive floating badge appears using CSS Anchor Positioning—click it to translate dialogue directly inside the artwork with zero copy-pasting.

• 🖱️ Hover: Translate instantly by hovering over any comic panel.
• 📌 Persistent: Keep translate badges visible on all comic images.
• 🚀 Auto-Translate: Automatically translates new panels as you scroll down the page.
• ✂️ Screen Crop: Translate protected canvas elements, WebGL games, or custom crops via right-click or Ctrl/Alt + Shift + X.
• ⚡ Reading Hotkeys: Press F to instantly peek at the original raw image, Ctrl/Alt + Shift + Z to translate the hovered panel, and Ctrl/Alt + Shift + M to open the chapter reader.
• 🎯 Right Click: Translate any image on the web via the browser context menu.

---

📖 Chapter Reader & Fast Scan

Read entire manga or webtoon chapters in one distraction-free view with automatic batch detection, continuous vertical webtoon mode, multi-page translation, and offline ZIP export:

• Fast Scan: Automatically auto-scrolls and queues all lazy-loaded manga panels across chapter pages.
• Flexible Reader Modes: Seamlessly switch between Webtoon continuous vertical strips, Single page, and Double spread layouts.
• Batch Chapter Translation: Translate the entire chapter in one click with progress tracking and offline preservation.
• 1-Click Chapter ZIP Export: Package all translated pages into a clean, numbered ZIP archive for offline reading and archiving with zero server uploads.

---

🛠️ Built-in Kites Studio

Need to polish a scanlation or adjust phrasing? Open the full-screen Kites Studio straight from the extension.

• Direct Inline Editing: Double-click any speech balloon to edit dialogue directly in place with subpixel vertical centering and live typesetting reflow.
• Interactive Canvas: Drag and resize speech bubbles with 8-point handles.
• Localized AI Inpaint Brush: Erase artwork defects, speech bubbles, or untranslated SFX with brush stroke precision.
• Side-by-Side Verification: Compare original, cleaned, and typeset views with one click.
• Typography Control: Live font sizing, colors, alignment, and line height adjustments.
• 1-Click Export: Download clean, translated pages at original resolution.

---

🎛️ Extension Controls & Presets

Customize target languages, neural OCR models, and inpainting algorithms from the clean popup:

• Custom Glossary: Define character names, honorifics, and specialized terminology to guide local and cloud LLM outputs.
• Site Domain Rules: Enable, disable, or assign custom element selectors per domain.

---

🧠 Powered by Modern AI

Component	Options	Description
Translation	WebLLM (Local WebGPU) Cloud Shared Pool (100/day free) Google Translate (Unlimited Free) Custom API (OpenAI, Claude, Gemini)	Run unlimited local LLMs in-browser (Qwen 3.5, Qwen 3, Gemma 3, Llama 3.2), use unlimited Google Translate, access our shared cloud pool (100 free requests/day), or connect your own API key.
Inpainting	LaMa Manga (Neural) AOT-GAN (Fast Neural) Telea Diffusion (Instant) Simple Fill	Erases text strictly inside character boundaries to preserve comic art.
OCR	PaddleOCR v3 – v6	Industry-standard multilingual text detection and recognition with seamless 1D webtoon strip tiling.
Bubble Fit	Neural YOLO (WebGPU) Heuristic (Instant) Classic MST	Expands dialogue into natural balloon whitespace with tail severing, eliminating tiny fonts.
Color-Bubble	Per-Bubble Detection (WebGPU) Font Color Match (Instant) Balloon Wrap (Instant)	Detects individual bubble text color, matches font color to original, wraps translated text in original balloon shapes.

---

🚀 Quick Start & Installation

Option 1: Chrome Web Store (Recommended • 1-Click Install)

The official and easiest way to install Kites. Includes automatic background updates and Google Web Store security review:

1. Open Kites on the Chrome Web Store.
2. Click Add to Chrome and confirm the prompt.
3. Pin Kites to your browser toolbar for easy access.
4. Browse any manga or comic page (on Twitter/X, Pixiv, MangaDex, etc.) and hover over a panel to translate!

---

Option 2: Pre-Built Zip (Manual Unpacked — Zero Terminal Required)

For offline environments, manual testing, or loading specific releases:

1. Download kites.zip from the latest GitHub Release.
2. Unzip the downloaded .zip file into a folder on your computer.
3. In Chrome, open chrome://extensions/.
4. Enable Developer mode (toggle in the top-right corner).
5. Click Load unpacked and select the unzipped folder.
6. Pin Kites to your browser toolbar and start reading immediately!

---

Option 3: Build from Source (Developers)

# 1. Clone repository
git clone https://github.com/Unheat/Kites.git
cd Kites

# 2. Install dependencies & build extension
npm ci
npm run build

1. In Chrome, open chrome://extensions/.
2. Enable Developer mode (top-right toggle).
3. Click Load unpacked and select the generated dist/ folder.
4. Pin Kites to your toolbar and start reading!

---

🤝 Contributing & Documentation

We warmly welcome contributions from developers, translators, and comic lovers!

• 📖 Documentation Hub — Architectural diagrams, pipeline specifications, and runtime contracts.
• 📐 Interactive Pipeline Map — Standalone interactive visualization of Kites' detection, inpainting, and typesetting algorithm.
• 🛠️ Contribution Guide — How to set up your dev environment, run tests, and submit PRs.

---

📜 License & Acknowledgements

Kites is licensed under the GNU General Public License v3.0.

Heartfelt thanks to the incredible open-source projects that made Kites possible:

• manga-image-translator (Cotrans): Groundbreaking spatial text grouping and typesetting algorithms.
• xianscan-rust: Dynamic patch inpainting architectures and OCR noise filters.
• koharu: Contour typesetting core — DP line breaking, contour font search, and hyphenation recovery ported 1:1.
• PaddleOCR: High-speed neural text detection models.
• @mlc-ai/web-llm: Browser-native WebGPU runtime for large language models.
• pyheal & OpenCV: The Telea Fast Marching Method inpainting implementation ported to TypeScript.
• InpaintWeb: Browser-side LaMa inpainting architecture reference.
• hypher & linebreak: On-demand Liang hyphenation patterns and UAX #14 line break opportunities.
• CRXJS (chrome-extension-tools): Manifest V3 Vite plugin powering the build toolchain.
• MTO Comic Fonts: Vietnamese scanlation-optimized community fonts bundled with the font library.

Full license texts, pinned revisions, and ported-functionality details: THIRD_PARTY_NOTICES.md.

---