9314JIUsi
🖋️ The Scribe Workshop · Text Restoration pdf.js self-hosted · Apache-2.0

PDF to Word: text restored, pages re-inked

Two ways to transcribe. Text mode stitches the PDF text layer back into paragraphs and headings by baseline, spacing and font size; scans automatically reroute to page-image mode, re-inking each page as a picture. Output is a real .docx — not a byte leaves your device.

Parsing engine on standby · first conversion downloads about 1.4MB from this site (pdf.js, cached for offline use)

01Manuscript Desk

Drop a PDF on the desk or click to browse · Ctrl/⌘+V paste works · ≤50MB

Current: text rebuild. Complex tables flatten into plain text, graphics are dropped, headers and footers may leak into the body — proofread after converting.

Downloading the parsing engine…

Your PDF and the generated docx are processed in this device’s browser only — nothing is uploaded, no account needed. The mode preference sits in localStorage and disappears if you clear site data. To wipe it now, use “Erase my data” in the footer.

02Scribing Desk

Progress 0/0 pages
Waiting for a manuscript
Nothing transcribed yet. Pick a mode and press Transcribe.
Rebuild preview

What this page can and cannot do

A PDF does not store paragraphs. It stores positioned text fragments — glyphs with coordinates. Turning those fragments back into editable Word paragraphs is a heuristic craft: fragments on the same baseline become a line; a line that ends far from the right margin, or a sudden indent on the next line, breaks a paragraph; a line set noticeably larger than the body text becomes a heading. This page runs pdf.js (Mozilla’s PDF engine, self-hosted here, ~1.4MB on first use then cached) plus a hand-written OOXML generator — entirely inside your browser.

Honest limits first. Text rebuild mode suits text-based PDFs (exported from Word, LaTeX, web pages): complex tables get flattened to plain text, graphics and charts are dropped, and headers or footers may leak into the body — it is a text-layer rebuild, not a layout clone. Page-image mode re-inks every page as a 150dpi picture inside the docx: the layout survives pixel-perfect, but not a single word can be edited. Scanned PDFs (almost no text layer per page) are detected automatically and routed to image mode — this page does not do OCR; if you need editable text from a scan, run it through an OCR service first.

The rebuilding rules (the scribe’s craft)

Facts

InputPDF (.pdf; password prompt for encrypted files; DRM ebooks not supported)
OutputWord .docx (text rebuild: editable paragraphs and headings; page images: full-page pictures, layout kept); named doc-text-rebuild.docx / doc-page-images.docx
Key optionsMode: text rebuild (default) / page images (150dpi); scans (under 5 text items per page) auto-suggest image mode; URL ?mode=text|image preset
ProcessingParsed and generated locally in the browser (self-hosted pdf.js, ~1.4MB once; the docx writer is built into this page); nothing uploads — only the mode preference sits in localStorage
Limits50MB per file; up to 100 pages converts directly, more warns but continues (page-by-page batching)
Known edgesText mode is a text-layer rebuild: tables flatten, graphics drop, headers/footers may mix in, 3+ columns degrade; image mode keeps layout but is not editable; no OCR; speed depends on your device

FAQ

Does PDF to Word upload my file?
No. Parsing and .docx generation both happen inside this device’s browser (the pdf.js engine is self-hosted on this site) — not a byte of your PDF leaves the device. The first conversion downloads a ~1.4MB parsing engine; after that it lives in the browser cache and works offline.
Will the converted Word look exactly like the original PDF?
No, and it would be dishonest to promise that. Text mode is a text-layer rebuild: paragraphs and headings are stitched by heuristic rules, complex tables get flattened into plain text, graphics and charts are dropped, and headers or footers may leak into the body. Page-image mode keeps the layout pixel-perfect but every page is a picture — nothing is editable. Proofread after converting.
Can a scanned PDF become editable Word text?
This page does not do OCR. Scans are auto-detected and routed to page-image mode: each page is rendered at 150dpi and embedded in the docx, layout preserved but not editable. For editable text, run the scan through an OCR service first.
Does it handle password-protected PDFs?
Yes, if you know the password: a prompt appears and the password stays in local memory only. DRM-protected ebooks and rights-managed files are not supported; the page reports that honestly.
What about two-column academic papers?
The page detects left/right column structure (bimodal x-coordinates) and rebuilds each column separately, left first. Three or more columns, or figures spanning columns, degrade into mixed text that needs manual cleanup.
Where are my settings stored?
The mode preference lives in this browser’s localStorage (key y9314:pdf-to-word-en:v1). PDF files and generated docx files never touch disk. The footer’s Erase my data button wipes the preference in one click.

Related

Need PDF pages as images instead? Use the PDF to JPG tool (72/150/300dpi); going the other way, merge images into a PDF with Image to PDF.