Free Japanese OCR, in your browser

A single Japanese sentence routinely mixes three writing systems — kanji for meaning, hiragana for grammar, katakana for loanwords — plus Latin letters and numbers, with no spaces between words. A model trained on English shreds that into noise. This page runs the Tesseract engine with its dedicated Japanese model:

• All three scripts are read together — kanji, hiragana and katakana in one pass, the way real text uses them.
• The artificial spaces are stripped. Raw Tesseract output breaks Japanese into segments with spaces ("会 議 は…"). This converter removes a space only when both neighbours are Japanese characters, so the text reads naturally while real spacing around embedded English words and numbers is kept.
• Everything stays on your device. The Japanese model is downloaded to your browser on first use — it's one of the larger ones, so the first conversion takes noticeably longer than later ones — and your image never leaves your machine.
• The output is Markdown, ready to paste into ChatGPT, Claude, a translation workflow or your notes, with a token count shown before you paste.

Horizontal text only. Books, magazines and traditional layouts often set Japanese vertically (tategaki, columns read right to left). This converter reads horizontal text; vertical passages come back scrambled. Screenshots, websites, chat apps, slides and modern documents are almost always horizontal and read well.

Furigana can add noise. The tiny reading hints printed above kanji in some books can be picked up as stray extra characters. Text without furigana converts cleanest.

Handwriting is out of scope. The model is trained on printed type; handwritten Japanese — especially cursive — mostly won't survive. Printed pages, screenshots and scans are what it's for.

Stylised fonts cost accuracy. Manga lettering, game UI fonts and decorative display type vary; plain body text in a standard font is the reliable case.