Manga OCR - Linux Setup and Thoughts

Manga OCR is a python-based OCR tool that works by detecting text in screenshots, copying the text to your clipboard, which you then can set up a text hooker to automatically paste to. It's a terminal application (aka runs without a GUI), and has become one of my most used programs when learning Japanese.

This guide exists because I struggled to set it up during my Linux testing. A lot of my issues were due to knowledge gaps, as well as installation not being well documented (or at least friendly to tech amateurs like me). But I felt instead of complaining about it being hard to figure out, I'd put something out there to help others. Writing with other newbies in mind, Page is subject to updates, clarifications, and fixes.

If you're on Windows & want to use manga-ocr, you can follow the old setup guide on LazyGuideJP. Just replace pip3 install mokuro with pip3 install manga-ocr (this was likely a typo, since the rest of the section refrences the right tool) I don't have access to a macOS device, so I can't comment on anything for there as of last updating.

Update 2026-04-15 - LazyGuideJP now recommends owocr, a tool that uses manga-ocr. I haven't tried it as of updating, but I'll leave the above note up anyways for archival sake. (Setup: Manga on PC - LazyGuideJP)

I'll also talk about my thoughts on the program after the set-up guide.

Manga OCR Setup On Linux

This guide is based around my experience setting it up in Kubuntu & Linux Mint (Debian/Ubuntu). Demo screenshots are taken in Linux Mint 22.2 in Virtual box.

Setting up Python

Make sure you have python 3.6 or newer installed. Some Linux distros have python pre-installed. You can check if you have it by running $ python3. If it drops you into python, you're good to go. If not, install based on your distro: (Downloading Python - Python Wiki)

You may also need to set up pip3, python's package manager in order to install python applications from the repos.

• $ apt install python3-pip

Installing Manga OCR

There's some extra steps you might have to take here, but I'll start off with the install command.

• Install Manga OCR: pip3 install manga-ocr

If you're on lower-spec hardware, make sure you have any unnecessary programs closed while installing. I'm not sure the exact reason, but something here took up a lot of memory & nearly crashed my old laptop running Kubuntu 25.04 on 8 gigs of ram. So I'll just recommend closing unnecessary programs & leaving as much ram for install as you can.

On Debian/Ubuntu based systems (like Linux Mint), If you just run it like that, you'll run into this error message:

This is because by default, the OS won't let you install python packages system-wide in order to reduce the risk of system breakages, and you're encouraged to set up a venv (python virtual environment). For a quick summary a venv lets you run a python project as a separate application independent from your system's main python install. (venv - Python Docs), (Stack overflow answer I refrenced)

Two ways to do this is to either use pipx, or making a venv yourself

venv with pipx

The easier option. Pipx is a tool that automatically creates a venw for python applications when you install them with it.

• Get pipx: $ apt install pipx
• pipx install manga-ocr
• pipx ensurepath will let you run manga_ocr (or other python tools installed this way) globally, like a regular command

Making your own venv

How I did it when I was testing kubuntu. It's faster to install with pipx, but I'll note this down for brevity.

You may need to install another python3 package to make a venv

• If you don't have venv support, you'll need to add it. For Debian/ubuntu: $ apt install python(version)-venv
• make a directory to host your venv and run $ python -m venv /your-directory/ to build it.
• Activate the venv: $ source /your-directory/bin/activate
• Install Manga OCR: pip3 install manga-ocr

Running Manga OCR

For reading images off your clipboard

• Run manga_ocr
• I don't use this one, as it'll add clutter when copy-pasting images you don't want OCR'ed, like context images for anki cards.

For reading images from a screenshot folder

• Run manga_ocr "/path/to/screenshot/directory/"
• Watches the selected folder & reads text off new images added to it. My preferred way, since by only reading from the folder you can still copy images you don't want OCR'ed without the program trying to read it.

If you installed it by making your own venv, you'll need to activate it each time you want to run Manga OCR

On initial start-up it'll take longer to boot as it needs to download the model (about ~400mb). But after that, it'll boot faster & can run completely offline.

Manga OCR is a terminal application, so it'll run in the background & copy recognized text to your clipboard until you close the terminal.

Example setup. Sample media is Pepper and Carrot episode 38. Install success!

Other Tools

Other things important to set-up

Clipboard

For reading images off your clipboard, and functionality with some texthooker tools, you'll need to install some clipboard tools. This differs depending on if you're using a wayland or X11 session.

• Wayland: Use wl-clipboard. This adds the utilities wl-copy and wl-paste, that lets you copy data between your clipboard & tools that rely on them.

• X11: Use xclip. Similar to wl-clipboard, this adds functionality like copying & pasting data between your clipboard & tools that rely on them.

Texthooker

You'll also need something to paste your text into:

• Anaceron DJT's texthooker page - HTML texthooker. Also has a downloadable offline version.
• Yomitan - Browser pop-up dictionary to use with the above. Also has it's own clipboard monitor page, but it only stores one sentence at a time.
• LAP Clipboard inserter (Firefox) - Automatically pastes your clipboard content when activated. This only runs when you activate it in a browser page.
• Clipboard Inserter Redux (Chrome)- I don't use chrome, but seems to be the one recommended for chromium based browsers

Screenshots

Some more robust screenshot tools

• Flameshot (works best on X11)
• Spectacle (KDE's default screenshot app. Works well with Wayland)

Thoughts on Manga OCR

Now that we're past the set-up guide, here's some notes on my experience using it.

My main use case for this is being able to do lookups & make Anki cards from images, screenshots, and hard-subs.

I like it because it's easier to run specific blocks of text vs a whole page, and I can run it offline if I need to. I also find it runs faster than some of the other tools I've tried. (I'm pinning this on being a non-graphical program). Like the name implies, it's also optimized for reading manga / vertical text, though it handles horizontal text pretty well too.

Despite this being a tool I like using, there are a few quirks that I'd like to let people know about:

Works Best with

• Large / isolated text
• Clear text
• No busy backgrounds
• Clean fonts / printed text.
• Smaller text blocks
• Vertical text

Struggles with

• Smaller text
• Blurry text
• Busy backgrounds
• Handwritten text (neater handwriting tends to do better)
• Multiple text blocks at once (Best to screenshot one block/bubble/line at a time)

My set-up for using it on both Linux and Windows is the same and relies on multiple desktops. These screenshots are taken on Windows 11 (sorry), but the only major difference between my setups is the screenshot tool I use.

Desktop 1 - Main Worktable

• Media I'm mining from
• Texthooker in browser + Yomitan connected to Anki
• Optional: Obsidian if I need to take notes for any reason (eg: noting timestamps, pages I left off at, story notes, or noting learning habits)

Desktop 2 - Anki & Unwanted Screen Clutter

• Anki, open to the card browser & connected to Yomitan
• Terminal (minimized, maximized here for demonstration)

Overall, despite it's quirks it's a pretty neat terminal application if you're willing to go through the set-up.

There's also some tools that use Manga OCR. Notably;

• Mokuro - Tools that processes batches of images & makes the text selectable. Useful for multi-page things like manga.
• YomiNinja - Overlay that lets you OCR your entire screen & make it text-selectable, with support for Yomitan & other pop-up dictionary plugins built in. Manga OCR's one of the models available, but I found it works poorly here compared to the others available.
• owocr - Similar OCR tool to YomiNinja, but more minimalistic. I haven't tested this one personally, but it's what the LazyJPGuide recommends in recent updates.

Revisions

• 2025-10-28 - Removed note on install size, since that was an overestimate based on my VM test. Still, it takes up a decent amount of disk space (5-10gigs)
• 2026-04-15 - Edited note about installing on Windows, as the LazyGuideJP now recommends owocr. But I also feel it's worth keeping the note up as the lack of guides for manga ocr on it's own is why this page exists to begin with. (unfortunately lost the archived page w/o the typo though... so that note goes unfixed 🥀)