pdfpix

Turn a Word document into a PDF

or drop one here
  • Text, fonts, bold, italic, sizes, colour, alignment, indents and line spacing carry over.
  • Tables, lists, footnotes, endnotes, headers, footers, hyperlinks and pictures carry over.
  • Shapes, text boxes, charts and SmartArt are drawn and counted in the result.
  • A legacy .doc file is refused. Save it as .docx in Word first.
Choose a Word document first.

Convert a Word .docx into a PDF without uploading it. This tab opens the OOXML parts, measures text against real font metrics, and draws every character as selectable text. Tables, lists, footnotes, headers, footers, links, pictures, shapes, text boxes, charts and SmartArt carry over, and Greek, Hebrew, Arabic and Chinese are drawn. The finished PDF is reopened and 98% of its text must come back.

How to turn a Word document into a PDF

  1. Choose a Word document

    Press Choose a document, or drop one on the page. The .docx is opened in this browser tab.

  2. The tool reads the package

    It opens the OOXML parts, resolves the styles a paragraph inherits, and collects the pictures and links the document refers to.

  3. It lays the pages out

    Text is measured against the real font metrics, broken into lines, and placed on pages with the section margins the document asked for.

  4. Check and download

    The new PDF is reopened with a separate reader and its text compared with the text the layout used. A failed check does not produce a download.

What the conversion can preserve

A .docx describes a document as flowing content plus the rules for laying it out. A PDF describes finished pages. This tool does the layout the word processor would do: it resolves the styles each paragraph inherits, measures every run against the metrics of the font the document names, breaks lines, fills pages, and places tables, footnotes, headers, footers, links and pictures where the section geometry puts them.

Shapes, text boxes, groups, charts and SmartArt are drawn as vectors, with their fill, their outline, the angle a rotated shape sits at, and the text inside them still selectable. An equation is typeset as the line a person would write by hand. Word keeps a formula as a tree, so the characters in file order give a+b2 for a plus b over 2, and the one fact that made it a formula is gone. This tool walks the tree instead and writes (a+b)/2, raises an exponent rather than writing a caret, and adds brackets only where the two-dimensional layout carried the meaning. What is left is an embedded object another program owns, such as a spreadsheet dropped into the page. Those are counted by kind and the count appears on the receipt, so a page that lost one says so. Legacy binary .doc files are refused at the first four bytes rather than partly converted.

Hebrew, Arabic and Chinese are laid out in their own direction rather than refused. A right-to-left paragraph is resolved character by character, so a Latin word or a run of Western digits inside it still reads left to right, while Arabic-Indic digits keep the direction the surrounding text gives them. Arabic letters are joined into their initial, medial and final shapes before they are drawn. Chinese breaks between characters, since the line has no space to break at, and a closing bracket or full stop is never left at the start of a line.

Greek, Cyrillic and the characters a formula is built from are drawn by the same route: the operators, the arrows, the relation and set symbols, the letterlike symbols, the technical and geometric shapes, and the bold and italic alphanumerics that sit outside the ordinary Latin letters. Before these faces existed a maths paper converted to a page with holes in it and was refused by the text check.

The face comes from the document when the .docx embeds one; when it does not, this site serves subsets of Noto Sans, Noto Sans Math, Noto Sans Hebrew, Noto Sans Arabic and Noto Sans SC from its own origin, and the tab fetches only the parts that hold the characters on your pages. A page of ordinary Latin prose fetches none of them.

The check is the reason there is a download. The PDF this tool wrote is reopened with pdf.js, a reader that shares no code with the writer, and its non-space characters are counted against the characters the layout consumed. Below 98% nothing is offered.

Questions

Does Word to PDF upload my document?
No. The .docx is opened, laid out, drawn, reopened and checked in this browser tab. The bytes never leave it.
Does it accept a legacy .doc file?
No. A .doc is the old binary format, not a zip of XML parts, and this tool refuses it instead of guessing. Open it in Word and save it as .docx first.
What carries over?
Text, fonts, bold, italic, sizes, colour, alignment, indents, line spacing, tables, lists, footnotes, endnotes, headers, footers, hyperlinks, pictures, shapes, text boxes, charts, SmartArt and equations.
How is an equation drawn?
As the line a person would write by hand. Word keeps a formula as a tree, so a fraction holds a numerator and a denominator and nothing in the file says one sits over the other. This tool walks that tree and writes the inline form: a plus b over 2 comes out as (a+b)/2, an exponent is raised and reduced instead of written with a caret, a square root gets its sign, and brackets are added only where the two-dimensional layout carried the meaning. Every character you typed is counted against the 98% floor. The brackets and signs the tool adds are drawn and never counted.
Does it handle Hebrew, Arabic, Chinese, Devanagari, Thai, Greek or maths symbols?
Yes, and 25 scripts in all. Arabic letters are joined and Hebrew and Arabic lines are laid out right to left, with any Latin or digits inside them running the other way. Chinese wraps between characters, and a closing bracket or full stop is kept off the start of a line. Thai, Lao, Khmer and Myanmar write a sentence with no space in it, so the line breaker asks the browser for the word boundaries. Greek, Cyrillic, the operators, the arrows and the letterlike symbols a formula is built from are drawn as well, so a physics paper converts instead of being refused. The font comes from the document first: a .docx that embeds its own face is drawn with that face and fetches nothing. Otherwise this site serves Noto subsets from its own origin, and only the parts your characters fall in are downloaded.
What is left out of the PDF?
Embedded objects another program owns, such as a spreadsheet dropped into the page, and pictures the document links to instead of storing. The receipt counts every item that was dropped rather than hiding it.
Will the pages break in the same places as Word?
Usually, not always. Line breaking uses the metrics of the font the document names, so a page whose font is not present on this machine can break differently. The text itself is unchanged.
Is the PDF text selectable, and what does the check prove?
Every character is drawn as text with its own font, never as a picture of a page. The tool then reopens the PDF it wrote with pdf.js and counts the non-space characters it finds. At least 98% of the text the layout used must come back, or no download is offered.

The other pdfpix tools

All 33 of them run in a browser tab like this one. Each reads its own output back and checks it before offering the download, and none of them needs an account.

Where to go next