Convert Japanese PDF to Markdown without broken spacing
Turn digital or scanned Japanese PDFs into structured Markdown with selectable Japanese OCR. Keep headings, tables and formulas, understand the real limits of vertical text and furigana, and prepare a reviewable source for search or RAG.
Choose Japanese OCR, then review the details that matter
A Japanese PDF can contain real text, page images, or both. The converter reads an existing text layer when it is trustworthy and uses OCR for image-only pages. Choose 日本語 in the OCR language selector so the engine loads the right recognition model. MinerU is the default and the stronger choice for Japanese scans in our verification; Docling remains useful for clean, simple documents. The output is ordinary Markdown that you can preview, copy, download and validate against the original PDF.
Convert a Japanese PDF in the web app
No account is needed for the free browser flow. Your OCR choice applies to this conversion and is remembered locally.
Add the PDF
Open the web converter, upload a file, drag it onto the page, or paste a direct PDF URL. Digital, mixed and image-only documents use the same flow.
Select 日本語 for OCR
Set OCR language to Japanese. Automatic OCR handles image-only pages. Enable Force OCR when selectable text exists but copying it produces mojibake, false characters or missing lines.
Convert with MinerU first
MinerU is the default because it was more accurate on the Japanese fixture and handles complex scans well. If a clean digital layout looks unusual, compare the Docling result rather than assuming one engine wins on every file.
Review and export
Check names, dates, amounts, tables, vertical passages and ruby annotations in rendered and Raw views. Then copy the Markdown or download the .md file.
No word spaces is normal; false spaces are not
Japanese typography carries structure differently from English. A reliable workflow keeps that distinction instead of treating every unusual gap or width as an OCR error.
No word boundaries
Japanese normally joins kanji, hiragana and katakana without spaces. Do not add English-style gaps.
False PDF spaces
Glyph positions and OCR boxes can insert gaps inside names or compounds. Check Latin text and numbers before removing them.
全角 and 半角
NFKC can align full-width and half-width forms for search, but it is not a lossless repair.
Mixed scripts
Product codes, URLs and Latin abbreviations need their spaces and punctuation preserved.
Dense tables
Invoices and forms use narrow columns. Verify row alignment and totals even when the prose is accurate.
Small annotations
Furigana and footnote marks are useful, but among the least reliable OCR elements.
What our Japanese OCR check actually showed
The EP-070 production verification used image-only Japanese fixtures on both conversion pools. These findings describe a controlled check, not a guarantee for every font, scan or historical document.
Horizontal Japanese
縦書き and furigana
Use the source as evidence. For contracts, invoices, personal names, dates and amounts, keep the PDF beside the Markdown and review every value that drives a decision.
Normalize a copy, not your only source
OCR output should remain auditable. Save the original Markdown first, then create a normalized derivative for retrieval, embeddings or full-text search. This lets you improve matching without hiding what the converter returned.
Preserve the raw conversion
Store the original Markdown with its source file name and conversion date.
Apply NFKC deliberately
Match ＡＢＣ１２３ with ABC123, but review compatibility symbols and meaningful typography.
Repair false spaces conservatively
Join obvious Japanese compounds while preserving spaces around Latin words, model numbers, URLs and units.
Chunk by document structure
Use headings, paragraphs and tables before fixed token windows. Add source and page metadata for citations.
# 請求書 ## 請求先 株式会社サンプル | 項目 | 数量 | 金額 | | --- | ---: | ---: | | API利用料 | 1 | ¥12,000 | 出典: invoice-2026-08.pdf / page 1
One workflow for common Japanese documents
請求書, 契約書, 論文 and 官公庁資料 need different checks: invoice totals and rows; contract names, dates and clauses; paper formulas and reading order; government tables, era dates and older typography.
Browser workflow
Use the free Web converter for interactive review. Japanese UI selects Japanese OCR by default unless another language was saved.
REST API workflow
Pass ocr_lang inside per-job options. The accepted language codes are ru, en, ja, es, pt, de and fr. Omitting the field preserves the server default.
curl -X POST https://pdf2md.dev/api/v2/jobs \
-H "Authorization: Bearer p2m_…" \
-H "Content-Type: application/json" \
-d '{
"url": "https://example.com/japanese-report.pdf",
"options": {"ocr_lang": "ja", "engine": "mineru"}
}'
The option is per request. It does not change account defaults, and older clients that send no ocr_lang continue to use the existing behavior.
Ready to convert a Japanese PDF?
Start in the browser, or use the same Japanese OCR pipeline from code and AI agents.
Japanese PDF OCR questions
Can OCR convert a scanned Japanese PDF to Markdown?
Yes. Select Japanese OCR. MinerU was near-perfect on the controlled fixture, with minor character drops.
Why are there spaces inside Japanese words?
PDF positions or OCR segmentation can create false gaps. Preserve intentional spaces around Latin text and numbers.
Does Japanese OCR handle 縦書き?
With caveats. Reading order was correct, but punctuation, line breaks and occasional characters were lost.
Will furigana be preserved?
Not reliably. Small ruby may be merged, detached or omitted, so check important readings manually.
Should I use MinerU or Docling for Japanese?
Start with MinerU. Docling was readable but weaker and is useful as a comparison.
Is Japanese OCR available on the free plan?
Yes. Both pools passed verification; normal free limits still apply.