Review
manga-image-translator review: how to use it, and is there an online version?
manga-image-translator is a free, open source Python program (GPL-3.0) that finds, reads, erases and translates the text in manga and other images. You run it on your own computer with pip or Docker, in batch, web or API mode. Its hosted version, Cotrans, had no running worker on 20 November 2026. It has no page editor, and its README says it is still early in development.
Table of contents
At a glance
| Developer | zyddnys and contributors on GitHub. Started February 2021. 10,623 stars and 1,086 forks on 20 November 2026source |
|---|---|
| Price | Free, GPL-3.0. Online translators you connect, such as DeepL or OpenAI, bill you on their own termssource |
| Install | Python 3.12 or later with pip (a C++ compiler for one dependency), or the Docker image of about 15 GB. No installer, no phone appsource |
| Ways to run | Batch mode on a folder, a web page on your own machine, an API, or a command line through Dockersource |
| Hardware | CPU by default. The --use-gpu flag uses CUDA or Apple MPS with a PyTorch build that matches your systemsource |
| Languages | "Mainly supports Japanese", plus Chinese, English and 20 other languages. 25 language codes in the referencesource |
| Translators | Online with your key (DeepL, ChatGPT, Gemini, DeepSeek, Groq, Baidu, Youdao, Caiyun and more) or offline models (Sugoi, NLLB, M2M100, Qwen2 and more)source |
| Output | PNG, WEBP or JPG. XCF, PSD or PDF through GIMP, which turns text into image layers in a PSDsource |
| Online version | The README links touhou.ai/imgtrans, which now opens Cotrans. On 20 November 2026 the Cotrans status API showed 0 workers and 30 images waitingsource |
| Activity | Last commit 18 November 2026. Last tagged release beta-0.3, April 2022source |
manga-image-translator is the most starred open source manga translator on GitHub, with 10,578 stars on 8 November 2026. The Cotrans web service is built on it. This review covers what it does, how to use it step by step, whether an online version still runs, and when a different tool is the better choice. Every fact links to the project's own README or GitHub page.
What is manga-image-translator?
The project aims to translate images "that are unlikely to be professionally translated", in its own words, such as comics shared in group chats and on image boards. It runs a page through a chain of steps, and each step is a setting you can change:
- Detection finds the text lines. Options include default, ctd, paddle and craft (the README says not to use craft for manga).
- OCR reads the text, with 32px, 48px and Manga OCR models.
- Inpainting removes the source text and repairs the art. The README recommends lama_large. This is the step a cleaner does by hand (see manga cleaning).
- Translation uses an online service with your own key, or an offline model on your machine.
- Rendering sets the translation back in. The manga2eng renderer tries to fit the bubble rather than the detected text lines.
- Extras: an upscaler (waifu2x, ESRGAN, 4xUltraSharp) and a colorizer (mc2).
The README is honest about the state of the project: "This project is still in the early stages of development and has many shortcomings." Its list of future plans adds that the text area comes from the detected text, not the bubbles, so it "cannot perfectly perform English typesetting".
How do you use manga-image-translator?
These steps follow the README. Commands are as the README writes them.
- Install Python 3.12 or later. The README warns that the newest Python may not yet work with some PyTorch libraries. On Windows, install the Microsoft C++ Build Tools first.
- Clone the repository with
git clone https://github.com/zyddnys/manga-image-translator.gitand create a virtual environment. - For GPU speed, install the PyTorch build for your CUDA version from pytorch.org before the next step.
- Install the requirements with
pip install -r requirements.txt. Models download to./modelsthe first time they run. - Add API keys for any online translator to a
.envfile in the project folder, for exampleDEEPL_AUTH_KEY. - Translate a folder with
python -m manga_translator local -v -i <path>. The results go to<path>-translated. - Fix the text if needed. Run with
--save-text, edit the text file, then render again with--load-text.
If pip gives you trouble, Docker is the simpler path: the image zyddnys/manga-image-translator:main holds all dependencies and models, at about 15 GB. For a browser page on your own machine, start server/main.py and open port 8000. The same server gives an API, with its documentation at /docs.
What manga-image-translator does well
- It costs nothing and you can read the code. GPL-3.0, a Discord server and recent commits (18 November 2026).
- Batch work. Point it at a folder and it translates every page, with retries and an option to skip images without text.
- Full control. A JSON config file sets every option. Run
python -m manga_translator config-helpto see the schema. - Offline translation. Sugoi, NLLB, M2M100, mBART50, Qwen2 and JParaCrawl run on your machine, so text need not leave it. The README recommends Sugoi for Japanese to English.
- Terms you set. A glossary file keeps names the same with the OpenAI translator, and pre and post dictionaries fix common OCR errors and mistranslations with regular expressions.
- Help for typesetters. The
--prep-manualflag writes blank, inpainted pages plus copies of the originals, ready for hand typesetting.
What manga-image-translator does not do
- No editor. To fix a line you edit a text file and render again. There is no canvas to move or restyle a bubble.
- No editable PSD. PSD export goes through GIMP, and the README says GIMP "will convert text layers to regular images" in a PSD. XCF keeps text layers.
- No sound effect step. Drawn sound effects stay in the source language unless you redraw them. Our post on why reader tools skip sound effects explains why this is hard.
- No easy setup. You need Python, a compiler, the right PyTorch build and, for online translators, your own keys.
- No dependable hosted version. See the next section.
Is there a manga-image-translator online version?
The README still lists an "official demo site" at touhou.ai/imgtrans and a browser script on Greasy Fork. The demo address now opens Cotrans, a web front end built on this project by VoileLabs. Our own check on 8 November 2026 found the site loading and its status API reporting 1 worker processing 6 images. On 20 November 2026 the site loaded, but the status API reported 0 workers and 30 images waiting. It runs on one worker, with no promise of uptime. Read our Cotrans review for what that service offers today.
manga-image-translator compared with other tools
| manga-image-translator | Koharu | Bublate | |
|---|---|---|---|
| What it is | Python program (README) | Desktop app (GitHub) | Web app in the browser |
| Price | Free, GPL-3.0 | Free, MIT or Apache 2.0 | $3 for a webtoon chapter up to 20 segments, up to $0.60 a manga page, per target language, 30 free credits |
| Editor | None, edit a text file | Canvas and inspector | Page editor with live text |
| PSD | Through GIMP, text flattened | Layered PSD with text layers | Layered PSD per page |
| Sound effects | No step | Not covered in its docs | Redrawn in the art, or kept with a caption |
| Setup | Python or Docker, your keys | Installer, local or hosted models | None, sign up and upload |
Bublate is the hosted tool we make, so weigh that row with care. For the two open source tools side by side, read Koharu vs manga-image-translator and the Koharu review. For every option, including browser tools, see the best manga translators. If you own the pages and want a finished release without the setup, Bublate does the first pass from $3 a webtoon chapter, up to $0.60 a manga page (pricing), and you can start with 30 free credits.
Who it fits
manga-image-translator fits someone who is happy in a terminal and wants to run hundreds of pages through a script, call a translator from their own code, or tune every step (detector, OCR, inpainter, translator) in a config file, at no cost. Pick something else if you want an app with an editor, a guaranteed online service, sound effects redrawn in the art, or lettering that fits each bubble the way a typesetter would.
Side by side, row by row: Koharu vs manga-image-translator: which open source manga translator should you use?
Frequently asked questions
How do you use manga-image-translator?
Install Python 3.12 or later, clone the repository and run pip install -r requirements.txt, or pull the Docker image. Then run python -m manga_translator local -v -i with the path to your folder. The translated pages go to a new folder next to it. Models download on the first run.
Is there a manga-image-translator online version?
Not one you can count on. The README links a demo at touhou.ai/imgtrans, which now opens the Cotrans site. On 8 November 2026 the Cotrans status page reported 1 worker processing images. On 20 November it reported 0 workers and 30 images waiting. For real work, run the program on your own computer.
Is manga-image-translator on Hugging Face?
The README links no Hugging Face Space. A search of Hugging Face Spaces on 8 November 2026 found two Spaces with that name from an unrelated account, with no likes. Treat copies like that with care, because your pages go to whoever runs them. The official code is on GitHub under zyddnys.
What is zydus/manga-image-translator?
It is a common misspelling. The project lives at github.com/zyddnys/manga-image-translator, with a double d and no u. There is also a separate Rust rewrite by another author, which its README says may be easier to set up because it is a compiled program, but it has only a command line.
Is manga-image-translator free?
Yes. The code is free under the GPL-3.0 licence, and the offline translators such as Sugoi, NLLB and M2M100 cost nothing to run. Online translators such as DeepL, ChatGPT, Gemini and DeepSeek need your own API key, and those services bill you for what you send them.
Does manga-image-translator translate sound effects?
Its README does not offer a sound effect step. It mentions sound effects once: a pre-translation dictionary can correct "irrelevant special effect text" before translation. Drawn sound effects in the art need a cleaner and a letterer, or a tool that redraws them, such as Bublate.
Sources
- manga-image-translator, README on GitHub (English) · checked
- manga-image-translator, GitHub repository (licence, stars, description) · checked
- manga-image-translator, commit history · checked
- manga-image-translator, releases · checked
- Cotrans status API · checked
- Hugging Face Spaces, search for manga-image-translator · checked
- manga-image-translator-rust on GitHub · checked