This site contains affiliate links. We may earn a commission at no extra cost to you.

How to Read Untranslated Doujinshi: A Practical Guide for English Readers

Most doujinshi will never be translated.

That is not a complaint about the industry — it is just arithmetic. Doujinshi are self-published. There is no publisher deciding which titles get a localisation budget. A work gets an English edition only if someone specifically chooses to translate that work, and there are far more works than there are translators.

When we counted one genre on DLsite, roughly one work in ten had an English edition available. The other nine were Japanese-only. Not obscure, not unpopular — simply not translated yet.

So if you read doujinshi in English, you spend a lot of time looking at works you cannot read. The useful question is not “where do I find translated doujin.” It is:

“What do I do about the nine out of ten that aren’t translated?”

This guide answers that. It covers what genuinely works, what looks like it should work but doesn’t, how to set your expectations so you aren’t disappointed, and — most importantly — how to tell in advance whether a specific work is worth the effort at all.

Part 1: Why the obvious solutions don’t work

Before the things that work, it’s worth understanding the thing that doesn’t. This is where most people lose an afternoon.

What you actually get when you buy

When you buy a doujinshi from a Japanese storefront, you don’t get a web page. You download a ZIP file. Inside it is a folder of numbered images — JPG or PNG, one file per page.

That is the whole product. There is no text file, no ebook format, no embedded subtitles. Just pictures.

This single fact explains everything that follows.

Why browser translation does nothing

Here is what almost everyone tries first:

• Right-click → “Translate this page”
• Paste the file into DeepL
• Turn on the browser’s built-in translation
• Try Google Translate’s document upload

None of these produce anything. Not a bad translation — no translation. The page looks exactly the same as before.

The reason is worth understanding, because it tells you what will work.

Translation tools operate on text. In a normal web page, the words exist in the page’s code as characters. The tool reads those characters, sends them off, gets different characters back, and swaps them in.

In a manga page, the Japanese is not text. It is drawn into the image. The characters inside a speech bubble are pixels, in exactly the same sense that the character’s hair is pixels. There is no character data anywhere in the file.

So when a translation tool looks at a manga page, it finds zero words. There is nothing to translate, and it correctly does nothing.

The takeaway: any tool that works on text will fail. You need something that can look at a picture and recognise letters inside it.

Why “just use an AI chatbot” is unreliable

A newer version of the same mistake: upload the page to a general-purpose AI assistant and ask it what the page says.

This sometimes works and sometimes doesn’t, and the failure mode is the dangerous kind. General-purpose models can read images, but they are not built for manga lettering — vertical text, hand-drawn styles, text that curves to fit a bubble, sound effects drawn into the artwork. When they struggle, they don’t always say so. They produce something fluent and plausible that isn’t what the page said.

A wrong translation you can detect is an inconvenience. A wrong translation you can’t detect is worse than none.

Purpose-built tools have the same weaknesses in principle, but they are tuned for this exact problem, and they show you the original alongside the translation so you can tell when something has gone wrong.

Part 2: What actually works — OCR

If the words are part of the image, you need OCR — optical character recognition.

An OCR translator does two jobs in sequence:

1. Detection. Find the Japanese characters inside the picture and work out where one line ends and the next begins
2. Translation. Translate what it found

Both steps can fail, and the first one is harder than people expect. Manga lettering is about as hostile to OCR as text gets: it runs vertically, it curves inside bubbles, it uses stylised fonts, and sound effects are often drawn by hand as part of the art.

This is why a tool built for manga usually beats a general OCR tool by a wide margin. It isn’t a better translator — it’s better at finding the text in the first place.

There are three practical approaches. They suit genuinely different situations, so it’s worth knowing all three rather than picking one and forcing it.

Approach 1: Phone camera or screenshots

Google Lens is the most widely available option. It’s built into most Android phones, and on iPhone it lives inside the Google app. It works on your camera view or on a screenshot you already have.

How to do it:

1. Open the page you want to read on any screen
2. Take a screenshot, or point your phone camera at the screen
3. Open Google Lens and select the image
4. Choose the translate function
5. The translation appears overlaid on the original

Where this shines: it takes about two minutes to try, requires installing nothing new for most people, and it works on absolutely anything — a screen, a printed page, a photo.

Where it hurts: you’re working one screenful at a time. For three or four pages this is fine. For a forty-page work it becomes tedious fast.

Use it for: finding out whether this whole approach suits you before you invest in anything. Also genuinely good for checking a single confusing page in a work you’re otherwise following fine.

Approach 2: Browser extensions built for manga

Several browser extensions exist specifically for translating manga. You open the image in your browser, the extension finds the speech bubbles, and it renders translated text over them — often clearing the original text out of the bubble first so the result reads naturally.

How to do it, in general terms:

1. Install the extension in your browser
2. Open your image files in the browser (most accept local files opened as tabs, and some accept a ZIP directly)
3. Trigger the extension on the page
4. Read; move to the next page and repeat, or let it process a batch

Where this shines: it’s the only approach that lets you read a whole work at something close to a natural pace. Because these tools are trained on manga specifically, detection is markedly better than general OCR — curved text and vertical writing cause far fewer problems.

Where it hurts: quality varies a great deal between tools, and it’s hard to know which is good without trying. Some are built primarily to work on specific reading sites rather than on files you own, which is a problem when your files came from a store.

Use it for: reading works front to back. If you’re going to read more than a handful of untranslated works, this is where you’ll end up.

Before you commit: test the extension on a work you already understand — ideally one where you also have the English edition, or one with very simple dialogue. That tells you what its output looks like when it’s working correctly, which makes it much easier to spot when it isn’t.

Approach 3: Desktop OCR tools

Screen-region OCR tools let you drag a box around any part of your screen and get a translation of whatever is inside it.

How to do it:

1. Open your images in whatever viewer you like
2. Trigger the OCR tool
3. Drag a box around one speech bubble
4. Read the translation
5. Repeat for the next bubble

Where this shines: it works on literally anything on your screen, including local files in any viewer. You also get precise control — you’re translating one bubble at a time, so you always know exactly which text produced which output.

Where it hurts: it is slow. Bubble by bubble through a whole work is a lot of dragging.

Use it for: works where precision matters more than speed, and as a fallback when an extension mangles a particular page and you want to check one line properly.

Comparing the three

If you’re starting from zero: try Google Lens today, and if you find yourself wanting to read whole works this way, move to an extension.

Part 3: What machine translation actually gives you

This section matters more than the tool choice. Most disappointment with OCR translation comes from expecting the wrong thing.

What you reliably get

The shape of the scene. Who is speaking. What they want. Whether something is being agreed to, refused, or negotiated. Where a scene turns.

For a great many works, that is genuinely enough to follow and enjoy the story.

What you reliably don’t get

Exact wording. Japanese-to-English machine translation is weakest in precisely the places doujinshi live:

• Casual speech and slang, which changes fast and doesn’t sit in training data neatly
• Dropped subjects. Japanese routinely omits who is doing something, because context makes it obvious to a Japanese reader. Machine translation has to guess, and it frequently guesses wrong — which means you can get the wrong character attributed to an action
• Sound effects, which are drawn as art and often aren’t detected at all
• Register and tone. The difference between someone speaking politely and the same person dropping into blunt speech is a real event in a Japanese scene. It usually vanishes entirely in translation

The failure that actually matters

The dropped-subject problem is the one to watch. If a line is ambiguous about who is speaking or who is acting, machine translation will produce a confident sentence with a subject filled in — and it may be the wrong one.

You won’t get an error message. You’ll get a clear, readable sentence that says something the page didn’t say.

Practical defence: when a line seems to contradict what the art shows, trust the art. The drawing is the original; the translation is a guess about the drawing.

When it’s simply not enough

If the entire point of a work is its writing — long conversations, a plot that turns on a specific line, wordplay — machine translation will flatten it and you will not know what you missed. That last part is what makes it worse than it sounds. You don’t get a warning that the good part didn’t come through.

For those works, waiting for an official English edition is a genuinely better plan than forcing it.

Part 4: Not every work needs translating at all

This is the section that saves the most time and money, and it’s the part most guides skip entirely.

Text volume varies enormously between doujinshi.

Some works carry almost everything in the artwork. The dialogue is a handful of short lines, the story is told by the sequence of images, and a reader with no Japanese at all can follow it comfortably. For these, OCR isn’t the solution — it’s unnecessary.

Others are the opposite: dense dialogue, long internal monologues, a plot that exists mainly in the writing. Those are exactly the works where machine translation fails worst.

So you’re really facing four different situations, and they call for four different decisions:

The problem: you can’t tell which one you’re looking at from the store page.

A product listing gives you a title, a page count, some tags, and a few sample images. It does not tell you how much of the work is talking. Page count doesn’t help either — a 50-page work can be nearly wordless or wall-to-wall dialogue, and the listing looks identical.

That gap is the reason this site exists. We sort works by how much Japanese you actually need, and we say why we put each one where we did. The reasoning matters as much as the label: a judgement you can’t check is just an opinion.

Part 5: How to judge a work before you buy

If you’re evaluating a work yourself, here’s what actually helps.

Read the sample images properly

Most listings include sample pages. Look at how much of each panel is covered by bubbles, not at what the bubbles say.

A page where speech bubbles take up a small fraction of the panel is a good sign. A page where bubbles crowd the art, or where you can see long unbroken blocks of text, tells you the work is carried by its writing.

Samples aren’t a random selection — they’re chosen to sell the work — but they’re still the best signal available on the page.

Watch out for text outside bubbles

Handwritten text squeezed beside the art, small annotations in the margins, and text integrated into the drawing are all much harder for OCR than clean text in a bubble. A work with a lot of this will translate badly even if the total word count is modest.

Page count tells you cost, not difficulty

A longer work isn’t harder to read — it’s just more of the same. What matters is text density per page. Use page count to judge how long the job will take, not whether it will work.

Check whether an English edition already exists

Storefronts are multilingual now, and the same work may exist as separate Japanese and English products. It’s genuinely common to buy the Japanese edition of something that had an English edition all along. Search the title before you buy.

Part 6: Official English editions, and why the catalogue keeps moving

Worth understanding before you reach for tools, because it changes the maths.

DLsite operates a translation programme in which creators authorise translations of their work, translators produce them, and revenue is split between them. Two details matter for you as a reader:

The translations are done by people. They are not machine output published as-is, and they are reviewed before release. That is why an official edition reads properly and OCR output doesn’t.

The catalogue is a moving target. New authorised translations appear continuously. A work that is Japanese-only today may well have an English edition in a few months.

The practical consequence: for a dialogue-heavy work you’re unsure about, waiting is a real strategy, not just giving up. For a light-on-text work, there’s no reason to wait — you can read it now.

Part 7: The buying step, briefly

One practical thing that catches people out at exactly the wrong moment.

Since April 2024, DLsite does not accept Visa, Mastercard or American Express directly. JCB works directly. For other cards, the route is through prepaid points — either points bought from official resellers, or prepaid codes.

It’s worth sorting out your payment method before you find something you want, rather than discovering the problem at checkout with a work in your cart.

This is a topic of its own and we cover it separately. Mentioned here only so it doesn’t ambush you.

Part 8: When things go wrong

Common problems, and what they usually mean.

Nothing gets detected at all. Usually the image resolution is too low, or you’re feeding the tool a compressed screenshot rather than the original file. Try the original image at full size.

Some bubbles translate, others are skipped. Normal. Skipped ones are typically vertical text, stylised fonts, or text that overlaps artwork. Fall back to a desktop OCR tool for those specific bubbles.

The translation contradicts what the art clearly shows. Almost always the dropped subject problem. Re-read the panel and trust the drawing.

Text comes out as fragments that don’t form sentences. Detection has split one line into pieces. If your tool lets you select a region manually, draw the box around the whole bubble and try again.

Sound effects are ignored entirely. Expected. They’re drawn as artwork. Some are worth looking up individually if a scene hinges on one; most aren’t.

Frequently asked questions

Do I need to know any Japanese at all? No. But knowing hiragana and katakana — a few hours of study — helps more than you’d expect, because it lets you sanity-check whether a name or a short line was read correctly.

Is it legal to run a translation tool on a work I bought? Translating something for your own private reading is a normal personal use. What you should not do is publish or distribute your translated version — that’s a different act entirely, and it’s not yours to distribute.

Can I just wait for everything to be translated? For popular works, often yes. For most works, no — there simply aren’t enough translators for the volume of works being published.

Which tool is best? The honest answer is that it depends on your platform and how much you read, and that the landscape changes. Start with Google Lens because it costs you nothing to try, and move to a manga-specific extension if you find yourself reading whole works.

Will translation quality improve? It has been improving steadily. The detection half of the problem — finding text in artwork — has improved more than the translation half.

Quick reference

• You get a ZIP of images. The Japanese is part of the picture, not text
• Browser translation and DeepL do nothing. Not badly — nothing at all
• OCR is the answer. Start with Google Lens; move to a manga extension for whole works
• Expect the shape of a scene, not the wording. Watch for wrong subjects; trust the art over the translation
• Check text density before buying. Sample images tell you more than page count
• Search for an English edition first. It may already exist
• Sort out payment before checkout. JCB direct; others via points

The single most useful habit: decide whether a work needs translating before you buy it, not after. Some don’t. Some aren’t worth it. Knowing which is which is most of the skill.

This site contains affiliate links. If you buy through them, we may earn a commission at no extra cost to you.

Related reading: Doujinshi With Almost No Text · Why OCR Fails on Manga · Does This Doujinshi Have an English Version? How to Check Before Buying

Where to start

Everything above is method. If you would rather start from works that need the least of it:

• Works where the art carries it
• Works a translation app handles
• Works with an english edition

Also worth reading: why ocr fails on manga