Local Manga Translator

🦊 Firefox Add-on Description & Instructions

⚠️ IMPORTANT: REQUIRES COMPANION SCRIPT This add-on is a "connector" tool. To perform the translation, you must run the free companion Python script (manga_server.py) on your computer.

✨ Key Features

• Universal Hardware Support: Works on NVIDIA (CUDA), AMD/Intel (Vulkan), or CPU.
• Specialized Japanese OCR: Uses Manga-OCR to read vertical, handwritten, and messy manga text perfectly.
• Advanced Bubble Detection: Now uses Comic-Text-Detector (specialized for Manga/Manhwa) to accurately split connected bubbles and ignore background noise.
• Smart Korean Mode: Uses PaddleOCR for high-accuracy recognition of Korean webtoons.
• Natural AI Translation: Connects to local LLMs (like Qwen, Llama 3) for human-quality translation.
• 100% Private & Free: No API keys, no monthly fees. Everything runs offline.

🛠️ Step-by-Step Installation Guide

Step 1: Install Visual Studio Build Tools (Windows Only)

1. Download Visual Studio Build Tools 2022 from the Microsoft website.
2. Run the installer.
3. CRITICAL: Select the workload named "Desktop development with C++".
4. Ensure the checklist on the right includes "Windows 10/11 SDK" and "MSVC... C++ x64/x86 build tools".
5. Click Install and wait for it to finish.

1. Download Python 3.10.11 from python.org.
2. CRITICAL: During installation, check the box "Add Python to PATH".

1. Get the Main Script

• Download manga_server.py from thehttps://github.com/Camekan/Manga_server.py/blob/main/manga_server.py and place it in a new folder (e.g., C:\MangaTranslator).

• Download this ZIP file: https://github.com/dmMaze/comic-text-detector/archive/refs/heads/master.zip
• Extract the ZIP. You will see a folder named comic-text-detector-master.
• RENAME that folder to: comic_text_detector
• ⚠️ Important: Use underscores _, not dashes -.
• MOVE this folder next to manga_server.py.

• The script will try to download this automatically. If it fails, do this:
• Download the model file: https://github.com/zyddnys/manga-image-translator/releases/download/beta-0.2.1/comictextdetector.pt
• Create a new folder named models inside your comic_text_detector folder if it isn`t there..
• Place the .pt file there.
• Correct Path: C:\MangaTranslator\comic_text_detector\models\comictextdetector.pt

MangaTranslator/ ├── manga_server.py └── comic_text_detector/ <-- The folder you renamed ├── inference.py <-- File inside ├── basemodel.py <-- File inside └── models/ <-- Folder inside └── comictextdetector.pt

Step 4: Install Dependencies

Open Command Prompt (cmd) inside the folder you extracted and run these commands:

1. Install Basic Tools:

pip install flask manga-ocr pytesseract pillow opencv-python numpy requests

2. Install Korean OCR (PaddleOCR):

pip install paddlepaddle paddleocr protobuf==3.20.3

3. Install AI Engine (Choose Your Hardware):

• Option A: NVIDIA Users (Best Performance)

• Option B: AMD / Intel Users (Vulkan Mode)
• Download and install the Vulkan SDK.
• Run this command:

• Option C: CPU Only (Compatible but Slower)

Step 5: Get an AI Model

1. Download a .gguf model (Recommended: Qwen2.5-14B-Instruct-Q4_K_M.gguf) from HuggingFace.
2. Open manga_server.py with a text editor (like Notepad) and set the MODEL_PATH to point to your downloaded file.

1. Double-click manga_server.py to start the server.
2. First run note: It will automatically download the detection model (~100MB). Wait for it to finish.
3. Open a manga page in Firefox.
4. Press Alt+Q (or your custom hotkey) and draw a box over the text!