# Comic Translate: AI-Powered Translation for Comics, Manga, and Manhwa > Comic Translate is an Apache-licensed Python application that uses large language models like GPT-4.1 and Claude-4.5 to translate comics, manga, manhwa, and other sequential art from and into multiple languages. It detects speech bubbles, runs OCR, inpaints the original text, and renders translated text in the cleaned bubble, all in a desktop GUI or browser extension. **ogkalu2/comic-translate** — AI comic and manga translator app/browser extension for automatically translating comics, manga, manhwa, BDs, fumetti, and more in multiple languages and formats (Images, PDF, EPUB, CBR, CBZ etc). - Repository: https://github.com/ogkalu2/comic-translate - Website: https://comic-translate.com - Stars: 2,964 · Forks: 343 - Language: Python - License: Apache-2.0 - Published: 2026-09-10 · Updated: 2026-09-10 · Language: en - Canonical page: https://hysenlabs.com/projects/ogkalu2-comic-translate ## The Translation Gap comic-translate Addresses The README states the problem directly: many automatic manga translators exist, and very few properly support comics of other kinds in other languages. Most existing tools were built for Japanese manga and do not handle Korean manhwa, French bandes dessinées, Italian fumetti, or Russian comics with the same quality. The quality argument for LLM-based translation is also explicit in the README: for distant language pairs like Korean-to-English or Japanese-to-English, large language models like GPT-4 outperform Google Translate, Papago, and DeepL by a significant margin. The README quotes from a specific work, 'The Walking Practice' by Dolki Min, as an example of where conventional translation devolves into incoherence that a SOTA LLM handles correctly. ## The Five-Stage Translation Pipeline The README describes five sequential stages. Speech bubble detection uses an RT-DETR-v2 model trained on 11,000 images of comics spanning manga, webtoons, and western formats, available on HuggingFace as `ogkalu/comic-text-and-bubble-detector`. Algorithmic segmentation then extracts text regions from the detected bubble boxes. OCR is the next stage. The default stack is language-specific: manga-ocr for Japanese, Pororo for Korean, PPOCRv5 for everything else. Two optional alternatives, Gemini 2.0 Flash and Microsoft Azure Vision, can be used for any supported language. Inpainting removes the original text. The project uses a manga and anime fine-tuned LaMa checkpoint and an AOT-GAN-based model. Translation uses the full page text fed to the LLM, with an option to provide the page image as additional context. The README lists GPT-4.1, Claude-4.5, and Gemini-2.5 as the supported translation models. Finally, text rendering places the translated text back into the cleaned speech bubble bounding boxes. ## Installing and Running Comic Translate from Source The desktop app and browser extension are available as downloads, but GPU acceleration requires running from source. The README provides these source setup commands: ```bash git clone https://github.com/ogkalu2/comic-translate cd comic-translate uv init --python 3.12 ``` Then install requirements: ```bash uv add -r requirements.txt --compile-bytecode ``` For users with NVIDIA GPUs, the README recommends: ```bash uv pip install onnxruntime-gpu ``` To launch the GUI from the comic-translate directory: ```bash uv run comic.py ``` This opens the desktop GUI. For updates, the README gives: ```bash git pull uv add -r requirements.txt --compile-bytecode ``` The project requires Python 3.12. The requirements.txt includes PySide6 for the GUI, pypdfium2 for PDF support, rarfile and py7zr for archive formats, and onnxruntime for the detection and OCR models. ## Supported Formats and Practical Usage Tips Comic Translate supports JPEG, PNG, PDF, EPUB, CBR, CBZ, and other image and archive formats. The supported source languages are English, Korean, Japanese, French, Simplified Chinese, Traditional Chinese, Russian, German, Dutch, Spanish, and Italian. CBR and RAR extraction requires a separate helper tool. The README names WinRAR, 7-Zip, Unar, and UnRAR as compatible tools. On Windows, the helper's folder must be added to PATH. On macOS, the desktop app checks standard Homebrew and MacPorts locations automatically when launched from Finder. The README describes several usage tips worth noting. Version 2.0 added a Manual Mode for situations where Automatic Mode fails: no text detected, incorrect OCR, or insufficient cleaning. Switching to Manual Mode allows corrections before translation proceeds. In Automatic Mode, processed images are loaded in the viewer while subsequent images continue processing in the background, allowing continued reading as the batch completes. Font selection affects output quality: the README warns that the selected font must support characters of the target language. ## Limitations: LLM Costs, OCR Accuracy, and CBR Extraction Every translation call sends page text, and optionally the page image, to a cloud LLM API. This means translation costs scale with the number of pages and the API pricing for GPT-4.1, Claude-4.5, or Gemini-2.5. The README does not provide cost estimates per volume. OCR accuracy depends on image quality and the specific OCR model. For Japanese, manga-ocr is specialized and generally performs well on clean manga scans. For other languages, PPOCRv5 is a general-purpose model that may struggle with highly stylized fonts or low-quality scans. The optional Gemini and Azure Vision alternatives can handle more cases but add API cost. The inpainting step, which removes original text, may leave visible artifacts on complex backgrounds. The README notes a Manual Mode specifically for cases where inpainting is insufficient. CBR extraction failure with the error `raise RarCannotExec("Cannot find working tool")` requires adding a supported extraction tool to PATH, which is a non-obvious setup step for users who encounter it. ## Desktop App Versus Browser Extension Comic Translate offers two distinct interfaces. The desktop application processes files: images, PDFs, EPUBs, and archive formats. It runs locally with all processing happening on the user's machine (except the LLM API calls). GPU acceleration requires building from source. The browser extension installs in Chromium-based browsers including Chrome, Edge, and Brave. It translates comics in-page on reading websites, removing the need to download image files. The extension is available through the comic-translate.com download page. The desktop app is also available as a pre-built download for Windows and macOS from the same site. The README notes a security override step for both Windows (Smart Screen) and macOS (Privacy and Security settings) when running the downloaded desktop app, which is standard for unsigned applications distributed outside official app stores. ## Maintenance and License The repository carries an Apache-2.0 license. The last push was on 2026-09-11, and version 2.8.9 was released the same day, with 2.8.8 released on 2026-08-10 and 2.8.7 on 2026-07-31. The release pace of roughly one release per month indicates active maintenance. The repository includes English, Korean, French, and Simplified Chinese README files, reflecting the multilingual user base. The project acknowledges several upstream dependencies: lama-cleaner for inpainting, manga-ocr for Japanese OCR, Pororo for Korean OCR, and the AnimeMangaInpainting checkpoint from dreMaz on HuggingFace. These upstream dependencies have their own maintenance trajectories, and an update to any of them could require a corresponding update to comic-translate. ## Conclusion Comic Translate is the right tool for readers who want to translate comics from languages without widely available official translations, particularly for East Asian formats like manga, manhwa, and webtoons where the LLM translation quality advantage over conventional tools is largest. The GPU acceleration limitation in the desktop app download, which is only available when running from source, is worth noting for users processing large volumes. The browser extension adds the ability to translate comics in-page on reading websites without saving files. Before setting up from source, confirm that your language's OCR model is covered: the default OCR stack uses manga-ocr for Japanese, Pororo for Korean, and PPOCRv5 for all other languages, with Gemini 2.0 Flash and Microsoft Azure Vision as optional alternatives. ## FAQ ### How do I translate a comic to English with Comic Translate? Download the desktop app from comic-translate.com, or run from source with uv. Open your comic file, select English as the target language, choose a translation model (GPT-4.1, Claude-4.5, or Gemini-2.5), and run in Automatic Mode. For CBR files, install a helper like WinRAR or 7-Zip and add its folder to PATH first. ### Does Comic Translate support manga? Yes. Japanese manga is the best-supported format: the speech bubble detector was trained on manga images and the default OCR uses the specialized manga-ocr model. Korean manhwa and webtoons are also explicitly supported, with Pororo as the default OCR for Korean. ### Is GPU acceleration available without building from source? No. The README states that GPU acceleration is currently only available when running from source, not in the pre-built desktop app download. Building from source uses uv to install the requirements including onnxruntime-gpu for NVIDIA GPU support. ## Sources - [License: Apache-2.0](https://github.com/ogkalu2/comic-translate/blob/main/LICENSE) - [ogkalu2/comic-translate on GitHub](https://github.com/ogkalu2/comic-translate) - [Project website](https://comic-translate.com) - [README](https://github.com/ogkalu2/comic-translate/blob/main/README.md) - [Releases](https://github.com/ogkalu2/comic-translate/releases) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/ogkalu2-comic-translate