← All guides

Can DeepL Translate a Scanned PDF? What Our Test Showed

DeepL is the strongest general translator we've run on scans. Here's what it did with four real scanned pages, what its own help center says about scans, and where it falls short.

6 min read

Yes. In our test DeepL read four image-only scanned PDFs, translated all of the text, and handed back PDFs with the original backgrounds in place. It is the strongest general-purpose translator we've tried on scans. It isn't flawless, though: on a page with text printed over a photo it left smudges and flattened the big numbers, and on a one-page research paper it returned two pages with the table split in half.

Below is what DeepL itself says about scanned files, what we saw when we ran real pages through it, and when a translator built for scans is the better pick. We make Reglyph, so we're not neutral. Every DeepL result described here is shown in full in our side-by-side test.

What does DeepL say about scanned PDFs?

DeepL's help center says you can translate PDF files with its file translation feature, on the web and in the desktop apps, and that you need a DeepL account to do it. Scans are mentioned directly: the quality of the scan might affect the quality of the translation, and so might custom fonts and large images.

Its troubleshooting article is more specific. It says PDF conversion relies on optical character recognition and may have a higher error rate. It recommends scanning at 300 DPI, and if a scanned file wasn't fully translated, it suggests scanning it again at a higher resolution. So DeepL supports scans by design, with a fair warning that a poor scan gives a poorer result.

What we tested

  • Four image-only scans: an annual report page with text over an aerial photo (English to Chinese), an 1871 French book page (French to English), a nutrition guide with colored panels (English to Spanish), and a financial table with a bar chart (English to Japanese).
  • Two digital research papers, for contrast: a physics paper (English to French) and a machine-learning paper with a parameter table (English to Korean).
  • Default settings, the same files and target languages for DeepL and for Reglyph. We judged layout and completeness, not the wording.
  • Six pages is a small sample. Read it as a close look at a few hard pages, not a benchmark.

Does DeepL keep the layout of a scanned PDF?

Mostly, yes. Each scan came back as a PDF of the original page with the translation set where the source text had been. The translation is real text, not a picture of text: in every file DeepL returned, we could select and copy it. DeepL never laid a solid box over the page to hide the original, and it never skipped a table.

On plain scans it is careful with detail. The 1871 book page kept its first-line indents and the italic book title, which our version lost. The financial table kept its bold headers and totals, with every number in its cell. The slips were small: a missing space in "frigateAbraham", a stray asterisk before the title, white blocks breaking the shading of one table row, and one bullet on the nutrition guide split across two columns mid-sentence.

Original
Reglyph(ours)
DeepLSlightly better
Jules Verne, Vingt mille lieues sous les mers (Hetzel, 1871), page 9. DeepL did slightly better here. Swipe to compare, tap to enlarge.

Where does DeepL fall short on scans?

Text printed over a photo

This was the hardest page in the set, and the clearest gap. Behind DeepL's Chinese, the briefcase and banknote icons are smeared where the English used to be, and the pie chart has a square patch cut into it. The three large display numbers became ordinary lines of text, and the 40% statistic lost the words "in coastal counties". Ours restored the icons and kept the numbers large, though in a thinner weight than the bold originals.

Original
Reglyph(ours)Better on this page
DeepL
USGS Woods Hole Coastal and Marine Science Center, 2024 Annual Report. Swipe to compare, tap to enlarge.

Figures and tables that have to stay whole

Our two digital papers aren't scans, but they show something worth knowing if your scan has diagrams or a table. On the physics paper DeepL translated the labels inside the main figure, and one of them came out several times too large and now covers the chart. On the one-page machine-learning paper it returned two pages: the parameter table moved to the second page and was split into two pieces, the figure captions stayed in English, and some configuration keys were translated, so hidden_dim became 숨겨진 차원.

What DeepL's help center tells you to watch for

  • Scan resolution: it recommends 300 DPI, and a rescan at higher resolution if a file wasn't fully translated.
  • Custom fonts and large images can lower the quality; it suggests documents with fewer visual elements where possible.
  • PDFs exported from CAD files, such as floor plans, aren't supported for file translation.
  • If the PDF was made from a Word or PowerPoint file, translating that source file usually gives a better result.
  • Text inside images in Word and Excel files isn't translated. Its advice is to convert the file to PDF first.

When is DeepL the right choice?

For a scan that is mostly text on a plain page, such as a letter, a typed report or a book page, DeepL is a very good choice. It keeps typography we don't, like first-line indents, italics and bold. If you already use DeepL for text, the file feature is right there, and on our digital papers its running text was the most polished of any tool we've tested.

When is a scanned-PDF translator the better fit?

When the background is a photo or a gradient, or when figures and tables have to come back whole and on the same page. That's where Reglyph did better in our test. You upload the scanned PDF as it is, and the translation is set back where the text was, so tables, stamps and numbers stay in place. It works in 40+ languages, and you can download the original and the translation side by side to check one against the other. It's machine translation, not a certified translation.

The surest test is one of your own pages. Your first 5 pages are free, with no credit card: try it here. For how DeepL ranks against ChatGPT, Gemini, Google Translate and four other tools on the same pages, see our PDF translation tools comparison.

Translate your scanned document now

Upload a scanned PDF — even one made with your phone's scanner app. Reglyph OCRs it, translates it, and rebuilds the page so tables, stamps, and figures stay exactly where they were.

Translate 5 pages free
No credit card · Files auto-delete in 24h · Never used for training

Frequently asked

Can DeepL translate a scanned PDF?

Yes. DeepL's help center covers translating scanned files and warns that scan quality affects the result. In our test it read four image-only scanned pages, translated all the text and returned PDFs with the backgrounds kept. It was weakest on a page with text over a photo, where it left smudges and flattened the large numbers.

Does DeepL keep the layout when it translates a PDF?

Mostly. On our scans the translation went back where the original text was, with first-line indents, italics and bold kept on the plainer pages. It slipped on figures and tables: an oversized label covered a chart, and a one-page paper came back as two pages with its table split.

Why didn't DeepL translate my whole scanned PDF?

DeepL's troubleshooting article gives two reasons: the file may exceed DeepL's size or character limits, or the scan quality may be too low. It recommends scanning at 300 DPI and rescanning at a higher resolution if a file isn't fully translated.

Is the text in DeepL's translated PDF selectable?

In our test, yes. Every PDF DeepL returned had the translation as real text that we could select and copy, laid over the original page.

Which is better for scanned documents, DeepL or Reglyph?

It was close in our test. Reglyph did better on a page with text over a photo, DeepL did slightly better on an 1871 book page thanks to its indents and italics, and the other two scans were ties. The full results are in our Reglyph vs DeepL comparison.

Translate a specific language

Pricing

Choose your plan

Simple, scan-friendly pricing. Pages are pages — no multiplier for scanned files.

Pay-as-you-go

Translate a few pages. No commitment.

one-time
pages

$0.30 a page, from 10 pages — you pay for exactly 10 pages.

  • Pages never expire
  • Up to 300 pages and 100 MB per file
Buy 10 pages — $3

Lite

Most popular

For individuals with scans to translate every month.

$9/ month
  • $0.09 / page
  • 100 pages every month
  • 5 bonus pages, refreshed daily
  • Up to 300 pages and 100 MB per file
  • Pages reset each month
Subscribe

Pro

For high-volume users with scans to translate every week.

$29/ month
  • $0.06 / page
  • 500 pages every month
  • 5 bonus pages, refreshed daily
  • Unused pages roll over (up to 1,500)
  • Up to 300 pages and 100 MB per file
Subscribe