Automatically translate a folder of Japanese manga pages to English with a single button click.
Use the offline Sugoi engine to translate manga without an internet connection or API key.
Manually fix imperfect text erasure in a translated page using the built-in inpainting brush.
Export translated comic pages to Word documents for further editing or proofreading.
GPU with at least 4-6 GB VRAM recommended for acceptable speed, a pre-built Windows release is available.
BallonTranslator is a desktop application for translating manga and comics. Given a folder of comic pages, it can automatically detect speech bubbles and text boxes, read the text, erase it from the image, run the translation, and then place the translated text back in the appropriate position. The whole process can be triggered with a single button click after you configure the source and target languages. The tool supports both Japanese manga and Western-style comics. The translation pipeline is made up of four independent steps: text detection, character recognition (OCR), background inpainting to cleanly remove the original text, and machine translation. Each step uses a separate model, and the end result tries to match the formatting of the original text, preserving things like font color, size, angle, and alignment. When translating into English or Chinese, layout adjustments are applied to help the translated text fit the bubble naturally. Several translation engines are supported, including DeepL, Sugoi (which works offline for Japanese-to-English), Caiyun, Papago, and Google Translate. Switching between engines is done through the settings panel. The tool also has manual editing modes: a canvas mode for painting over areas and an inpainting brush for fixing imperfect erasures, plus a text editor with basic formatting controls and find-and-replace across the whole document. Files can be imported and exported as Word documents. For Windows users, a pre-built release is available for download. Running from source requires Python and pip. GPU acceleration is supported if an Nvidia card is present, which the README recommends for acceptable processing speed. The GPU setup requires about 6 GB of video memory by default, with a setting to reduce that for cards with 4 GB. The project is primarily documented in Chinese, with README files in Russian, Japanese, and Indonesian also available. It builds heavily on a companion project called manga-image-translator, which provides several of the underlying detection, OCR, and inpainting models.
← dmmaze on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.