Japanese image to text is for getting Japanese out of a picture and into somewhere useful: a word from a textbook you want to look up, a line of game dialogue, a product label, a LINE chat, a menu, a manual, or a web page that shows its text as an image. Paste or drop the image in, check a few characters, and copy the result.
The page starts with Japanese and English selected, since Japanese text often includes English words, product names, units and romaji. Both are read in the same pass.
Three scripts on one line
Written Japanese mixes kanji, hiragana and katakana, often with Latin letters and numbers in the same sentence, in both normal and full-width forms (ABC123). The Japanese model covers all three scripts, so a sentence like 新しいスマホを買いました, with kanji, hiragana and katakana side by side, needs only the one language.
That mix also explains where many mistakes come from. Several katakana are almost identical to kanji or to each other, and the engine has to pick between them from tiny differences in stroke length and angle.
Characters that look alike
These pairs are easy to confuse in almost any font, so they’re worth checking first:
- Katakana and kanji twins: カ and 力, エ and 工, ニ and 二, ハ and 八, ロ and 口
- Katakana pairs: ソ and ン, シ and ツ, which differ in the angle of their strokes
- Hiragana and katakana: へ and ヘ look the same in most fonts
- Long vowel mark: ー, the kanji 一 and a dash
- Small kana: っ and つ, ゃ and や, ょ and よ, which differ only in size
- Voicing marks: は, ば and ぱ, where the difference is a tiny ゛ or ゜
If you paste a word into a dictionary and it isn’t found, one of these is a likely reason. Words the engine was unsure about are underlined, with a button that jumps from one to the next. Click any line and its place lights up on the image.
Vertical writing (tategaki)
Novels, newspapers, manga and many signs are set vertically, in columns read from top to bottom and right to left. The Japanese option here is for horizontal text, and there’s no vertical option in the language list, so vertical text isn’t something on-device reading handles well. It usually comes out wrong, or with the columns in the wrong order.
What you can do:
- Look for a horizontal version. A web or digital edition of the same text is often set horizontally, and that reads far better.
- Crop the horizontal parts. Headlines, captions and tables on a vertical page are often horizontal and can be read on their own.
- Try your phone for a few lines. For short vertical passages, the text recognition built into your phone, such as Google Lens or Live Text on an iPhone, is worth trying.
Furigana
Furigana, the small kana printed above kanji to show the reading, is common in children’s books, learner materials and manga. On horizontal text, it usually comes out as an extra line of small kana, or as stray characters next to the kanji it belongs to. Delete those lines in the editor, or crop them out when the furigana sits in a band of its own. Very small furigana may simply be skipped.
Games, manga and screenshots
Game screenshots with a horizontal dialogue box in a clean font are a good fit. Capture the game at its full resolution rather than photographing your TV, which adds glare and wavy patterns, and crop to the dialogue box so the background art isn’t read as text. Low-resolution pixel fonts are harder; enlarge the game window before capturing if you can.
Manga is a poor fit. The text is usually vertical, hand-lettered sound effects are part of the artwork, and furigana sits beside many words. Horizontal speech bubbles in webcomics have a better chance if you crop each bubble on its own.
Chats and web pages, such as LINE conversations or a site that won’t let you select text, are usually horizontal and read well. Zoom in before capturing small text, and leave dark mode on: Auto improve inverts light text on a dark background. The screenshot to text page lists shortcuts that copy a screenshot straight to the clipboard.
Spaces and line breaks
Japanese is written without spaces between words, but the engine works in units it calls words, and Image to Text App puts a space between those units. You may see spaces between characters that the original doesn’t have, and joining wrapped lines adds one at each join.
To clean them up, open find and replace with Ctrl+F (⌘F on a Mac). If the text has no English, replace a single space with nothing. If it does, tick regexp and search for (?<=[^\x00-\x7F]) (?=[^\x00-\x7F]), which finds only spaces between two Japanese characters, and replace all with nothing.
Getting a cleaner result
- Bigger is better. A kanji packs many strokes into a small square. Zoom in before taking a screenshot, or move closer for photos.
- Crop tightly. One paragraph, one dialogue box or one label at a time reads better than a busy full screen.
- Straighten photos. For a menu or sign shot at an angle, use four-corner straightening first. See photo to text.
For more general advice, read how to get accurate OCR results.
Where the text goes next
The result is ordinary Unicode Japanese, so it pastes straight into a dictionary app, a translator, a flashcard deck, Google Docs or Word. You can also download Word (.docx), plain text, Markdown or a searchable PDF, which keeps the original image with an invisible text layer you can search. A regular PDF download opens your browser’s print dialog for Japanese, where you choose “Save as PDF” so every character is drawn properly.
If the text turns out to be Chinese rather than Japanese, Chinese image to text starts with the right model.
Private by default
Your screenshots are read in your browser, on your device, and aren’t uploaded. History is off unless you switch it on, and then it stays in your browser. Read more in is online OCR private?