Comic translator that uses LLMs because DeepL turns manga into gibberish
It exists because most comic translators only handle manga, and off-the-shelf machine translation often devolves into gibberish on distant language pairs.

What it does
Comic Translate is a desktop app and browser extension that ingests comics, manga, manhwa, BDs, and fumetti—via images, PDF, EPUB, CBR, or CBZ—and produces translated versions. It detects speech bubbles with a custom-trained RT-DETR-v2 model, runs OCR with language-specific engines, inpaints the original text away, and renders new text in its place. The actual translation is handled by LLMs like GPT-4.1, Claude-4.5, or Gemini-2.5, which are fed the full page text (and optionally the image itself) to preserve context.
The interesting bit
The project treats translation as a generative rewrite rather than a word-swap, arguing that SOTA LLMs outperform Google Translate and DeepL for distant pairs like Korean↔English. It also explicitly targets Western comics and European BDs, not just manga, using a detector trained on 11,000 images spanning multiple comic styles.
Key highlights
- Supports source material in eleven languages including English, Korean, Japanese, French, Russian, and German.
- Uses a custom RT-DETR-v2 bubble detector and algorithmic segmentation to isolate text before erasing it with a manga-finetuned LaMa inpainting model.
- Feeds entire page context to LLMs rather than sentence-by-sentence strings, with optional image input for visual nuance.
- Includes a Manual Mode to correct automatic detection, OCR, or cleaning failures before final rendering.
- Reads PDF, EPUB, CBR, CBZ, and raw images.
Caveats
- GPU acceleration is only available when running from source; the pre-built Windows and macOS binaries appear to run on CPU.
- Opening CBR archives requires WinRAR or 7-Zip to be installed and available on the system PATH.
- The downloaded app triggers OS security warnings on both Windows (SmartScreen) and macOS, requiring manual overrides to install.
Verdict
Worth a look if you scan or read foreign comics in bulk and want a single pipeline that handles detection, cleaning, and contextual translation. Skip it if you need a lightweight, fully offline tool or if occasional manual correction of bubble masks sounds like too much friction.
Frequently asked
- What is ogkalu2/comic-translate?
- It exists because most comic translators only handle manga, and off-the-shelf machine translation often devolves into gibberish on distant language pairs.
- Is comic-translate open source?
- Yes — ogkalu2/comic-translate is open source, released under the Apache-2.0 license.
- What language is comic-translate written in?
- ogkalu2/comic-translate is primarily written in Python.
- How popular is comic-translate?
- ogkalu2/comic-translate has 2.8k stars on GitHub.
- Where can I find comic-translate?
- ogkalu2/comic-translate is on GitHub at https://github.com/ogkalu2/comic-translate.