Find the words you're studying, inside what you're reading

Go to the Library tab and click Load the example — or, if you've already got something in your library, use the small 3 manga chapters and beginner word list links above the buttons.

That gives you three chapters of a real Japanese manga and about 1,300 beginner words. Tick a chapter and the word list together and you'll see the coverage figures. Press Read this chapter and your words are highlighted on the page. Tap any word for its meaning, and Add to deck to make a card from it.

Delete any of it whenever you like — the links stay, so you can always put it back.

Everything — chapters, word lists, cards, review history — lives in your own browser, on the device you're using. There is no account, nothing is uploaded, and nobody else can see any of it.

The flip side: nothing is backed up unless you back it up, and clearing your browsing data will wipe it. On the Study tab, Back up saves a small file with your cards and your whole review history. Do it once you've built up a deck worth keeping.

Using a different device, or a different browser, means starting fresh — unless you carry that backup file across, which works fine.

A plain text file (.txt) or spreadsheet export (.csv, .tsv). The simplest version is one word per line:

食べる
図書館
静か

A reading and a meaning as extra columns make nicer flashcards:

食べる,たべる,to eat
図書館,としょかん,library

Commas, tabs or | all work as separators. A header row (word,reading,meaning) is understood but not required. Two spellings of one word go in a single cell separated by a slash — 夫/主人 — and both get looked for.

Verbs are best in dictionary form — 食べる rather than 食べます. Textbook lists often use the polite form, which won't match the plain and conjugated forms that actually appear in manga.

Japanese words change shape. Your list says 片づく; the manga prints 片付いた. Connecting those needs a Japanese dictionary far too large to live inside a web page.

So the desktop preparer (Part 3) works out every alternate spelling in advance and writes a .words.json file. Bring that instead of the raw .csv and you'll find noticeably more of your words. The example beginner list is prepared this way — it carries nearly a thousand alternate spellings.

A plain .csv still works. It just finds less, and the Library marks such lists with a ~ so you know which is which.

Python is a free program that other programs are built on.

Windows — get it from python.org/downloads.

This one matters: on the very first screen of the Windows installer there's a checkbox saying "Add python.exe to PATH". Tick it. It's off by default, it's easy to miss, and nothing works without it. If you already installed Python and things don't work, run the installer again and tick it this time.

Mac — open the Terminal app and run xcode-select --install, or use the python.org installer. No checkbox trap on this one.

Linux — sudo apt install python3 python3-venv python3-pip (or your distribution's equivalent).

Unzip the download somewhere sensible — not inside the zip viewer window, which is a classic way to get confusing errors. Then:

• Windows — double-click prepare.bat
• Mac — right-click prepare.command → Open → Open. macOS blocks double-clicked scripts the first time only; after that a normal double-click works.
• Linux — chmod +x prepare.command && ./prepare.command, or just python3 launch.py prepare_gui.py

A terminal window appears, then your browser opens on a page called Chapter preparer. That terminal window is the program — leave it open while you work.

The first launch is slow. It downloads the software that reads Japanese off images: around two gigabytes, once, ever. Expect 10–30 minutes. After that it opens in seconds.

Under 1 · your manga, click Browse and choose the folder holding your pages. It accepts:

• a folder of image files (.jpg, .png)
• a .cbz file
• a folder containing several of either — it does them all

Under 2 · options you can also choose a word list to prepare at the same time. Worth doing — see Part 2.

Use manga you own. Nothing is uploaded anywhere, but that doesn't make someone else's scans yours to pass around.

You'll see a choice of CPU or GPU.

CPU works on every computer and needs nothing extra. It is genuinely slow — roughly 2–5 minutes per chapter, so a long series is an overnight job.

GPU is many times faster but needs a modern NVIDIA graphics card, and downloads another large package the first time. If the button is greyed out, your computer can't use it — that's all it means, and CPU will still get you there.

Start it and go and do something else. You can stop partway and resume later: finished chapters are skipped, so nothing is wasted.

You get one .mmz file per chapter, plus a .words.json if you prepared a list. Those are the whole point — they hold the pages and the text together, they're much smaller than the originals, and they work on any device.

Back on this site, use Add chapters and Add word list. For a phone, put them in Dropbox or Google Drive and use the ☁ button, or just email them to yourself.

Tick a chapter and a word list together and you get two numbers, which answer two different questions:

• Your list covered — how much of what you're studying turns up here. Useful for deciding whether a book is worth reading for vocabulary.
• Running text covered — how much of the actual text on the page you'd already know. Useful for deciding whether it'll be a pleasant read or a slog. Below about 80% it's hard going.

Non-dictionary words are things no dictionary knows: character names, sound effects, invented words. They're counted separately because leaving them in makes a book look harder than it reads — you don't need to study ドキドキ to understand it.

The Aa button in the reader and in review turns on three optional things: furigana above the kanji, spaces between phrases the way beginner textbooks print them, and colour by word type. All three are off until you want them, and they're remembered per device.

The colouring is a grammar aid — useful for seeing how a sentence is put together. It won't make hard text easy; spacing and size do more for that.

The Study tab is your deck. Press Review and say how well you knew each word; when it comes back is worked out for you, and words you find hard return sooner.

Set a daily limit with the ⚙ button. The default of 20 new words a day sounds low and isn't — every new word comes back several times over the following weeks, so 20 a day becomes a real workload within a fortnight. Raising it is easy; digging out from under a 400-card backlog is not.

If you've added a lot at once, Sort unstarted goes through them quickly so words you already know don't waste your time.

Export to Anki is there if you already use Anki. Ignoring it costs you nothing — the review here is complete on its own. It is not a backup; see below.

Your cards and everything you've learned live inside this browser, not in an account somewhere. That's what makes this free and private. It also means nothing is backed up unless you back it up.

On the Study tab, Back up saves a small file with your cards and your whole review history. Keep it in a cloud folder or email it to yourself. Restore loads it back — on this device, a new laptop, or a new browser.

The file has no pictures in it, which is why it's small enough to email. After restoring, press Repair cards and the pictures come back from your chapters.

What can actually lose your work: clearing your browsing data, "reset browser", uninstalling the browser, or some privacy-cleaner tools. The site asks the browser to treat your cards as permanent, and most will — but clearing browsing data is you asking, so it wipes them regardless. A backup file is the only thing that survives that.

Restoring never deletes newer work: if a card exists in both places, whichever has been studied more is kept.

The example is Ubunchu! by Hiroshi Seo — real Japanese manga, shared by its author under Creative Commons Attribution-NonCommercial 2.1 Japan. Copying and adapting are allowed with credit; commercial use is not. His own licence page is page 2 of the first chapter, and the credit appears on the Library tab and under Credits below.

It's a comedy about three students setting up Linux, so expect コマンド and インストール rather than textbook vocabulary. Measured against the beginner list that comes with it: 201 of 1,327 words appear (15%), covering about 18% of the running text. That's low, and it's the real number — the example is there to show you the machinery working on real manga with real OCR, not to be your study material.

Which is rather the point of the coverage figures: they let you find out whether something is worth your time before you spend a month on it.