Word to PDF layout test: fonts, links, tables, and images

We rendered one controlled DOCX through Converttwo’s LibreOffice workflow and through Microsoft Word. Page count, searchable words, images, link, and page sizes matched; image placement and rasterized page pixels were not identical.

Controlled test result

Document structure survived, with small renderer-dependent shifts

Page sequenceThree portrait Letter pages in both PDFs
Searchable words64, 203, and 32 words on the same respective pages
Interactive linkOne URI link remained on page one in both PDFs
ImagesOne image on page one and one on page three in both PDFs

Controlled fixture and method

The 939,358-byte DOCX contains three sections, 23 paragraphs, one real Word table, two inline images, an external hyperlink, explicit section breaks, headers and footers, and text set in Arial, Georgia, Courier New, and Calibri-based heading styles. All text and artwork are generated for this test and contain no private material.

  1. Generate the source with scripts/evidence/generate_fixtures.py.
  2. Convert it through Converttwo’s current /api/convert/docx-to-pdf route.
  3. Open the same source in Microsoft Word and export a reference PDF.
  4. Inspect page geometry, searchable word counts, links, image boxes, and fonts with PyMuPDF.
  5. Render corresponding pages at 108 ppi and calculate changed-pixel and channel-error measurements.

The Converttwo run used application code at commit 3caf8a7 and LibreOffice 26.2.5.2 on Windows. The comparison reference used Microsoft Word 16.0.20326.20132 with its repair-on-open option enabled to prevent an interactive automation prompt; no intentional content edits were made. Production hosts can use different LibreOffice and font versions, so the environment is part of the result.

Page one retained the table, fonts, image, caption, and link

Microsoft Word reference, page 1
Converttwo output, page 1

Both files exposed 64 searchable words and one clickable link on page one. Both embedded Arial, Georgia, Courier New, Calibri, and their used bold variants. The main image’s 381.6 x 254.4 point dimensions were unchanged; its top moved from 199.31 points in the Word reference to 199.55 points in the Converttwo output.

The final image stayed on page three but shifted right

Microsoft Word reference: image begins 46.8 points from the left
Converttwo output: image begins 55.85 points from the left

The image kept its 468-point width and approximately 312-point height but moved 9.05 points to the right. The heading, explanatory sentence, final sentinel, and footer remained on the same page. This was the largest visible difference in the controlled fixture.

Measured output against the Word reference

Exact measurements from the two PDFs
CheckMicrosoft Word referenceConverttwo output
File size287,255 bytes297,784 bytes
Pages and dimensionsThree 612 x 792 point pagesThree 612 x 792 point pages
Words per page64; 203; 3264; 203; 32
Images per pageOne; zero; oneOne; zero; one
LinksOne URI link on page oneOne URI link on page one
Changed pixels at 108 ppiReferencePage 1: 3.70%; page 2: 6.14%; page 3: 22.74%
Mean absolute channel errorReferencePage 1: 1.827; page 2: 4.208; page 3: 3.540 on a 0-255 scale

Pixel differences include anti-aliasing and font-rendering changes, not only moved objects. The equal word counts, links, page dimensions, and image counts provide separate structural checks. Neither metric alone proves complete fidelity.

Downloads and limits

Limit of this test

This favorable fixture uses installed common fonts and inline images. It does not contain equations, tracked changes, text boxes, macros, embedded spreadsheets, unavailable fonts, or floating objects. Those features can produce larger differences. Microsoft Word was used only as a comparison renderer, not as part of Converttwo’s output path.

Convert the reviewed DOCX

Compare page count, links, fonts, tables, and object placement before distributing the PDF.

Open Word to PDF

Related evidence and guidance