Translate PDF

The layout stays put: each block is covered and rewritten in its own space, with a glossary you control and every segment reviewable before the file.

Open in PDF ARENA

The layout stays

Each block is covered and rewritten in its own space, with the type size measured to fit.

Your glossary wins

Terms you force are applied after translation, so the model cannot override them.

You review before it is written

Every segment side by side with its original, editable. Nothing is built until you say so.

Standard PDF fonts only cover Latin script with accents, so target languages are limited to those. hi, ru, zh, ja, ko, ar and other scripts need an embedded font file this tool does not carry, and offering them would return a document full of question marks.

Translation that keeps the page

Most PDF translators hand back a wall of text. Tables gone, columns collapsed, and the document you needed to send somebody is now a transcript. Here the page survives. Text comes out with its geometry, gets grouped into blocks that mean something, and each translated block is written back into the exact rectangle its original held. Type size is measured against that rectangle: run longer and the letters shrink, within a limit you set. Past the limit it is reported as an overflow rather than quietly spilling into the next paragraph. Blocks, not lines. A line cut mid-sentence translates badly in every language, because the translator cannot see where it ends.

How to translate a PDF and keep the layout

1

Open your PDF

Text and geometry are extracted on your device, then grouped into reading blocks.

2

Read the consent panel

What gets sent, to whom, what happens to it. Pick the target language and load a glossary if you have one.

3

Review every segment

Original above, translation below, editable. Terms your glossary forced are marked.

4

Build the document

Translated PDF, bilingual page by page, or just the segments as JSON.

Fitting, glossary and the honest limit

Blocks are cut where the vertical gap passes 1.7 line heights, where the indent jumps, or where the type size changes. Those three separate one paragraph from the next in a laid-out page. Fitting has one lever and one rule. The lever is type size, tried down in quarter-point steps until the wrapped lines fit. The rule is that horizontal squashing is never used: a block that will not fit at your floor is reported as an overflow, with its page, instead of deformed into place. The glossary lands after translation, never before. Substitute a forced term first and the model translates your substitution, which defeats the point. Afterwards, the glossary wins. An empty target keeps the term as it was: that is how you say do not translate this. Batches come back numbered, and each numbered line goes to the block that asked for it. When fewer lines return than were sent, the missing blocks stay empty and get counted. Filling them with the original would look like a translation and be a provider failure in disguise. The limit worth stating plainly: the fourteen standard PDF fonts cover Latin script with accents, nothing else. Hindi, Russian, Greek, Arabic, Chinese, Japanese, Korean, Polish and Czech need an embedded font file, and this tool carries none. So those languages are not offered. A character the font cannot write is counted and reported, and its block stays in the original language.

Why this one

Each block is rewritten in its own rectangle, with the size measured to fit.

Your glossary is applied after translation, so it always wins.

Every segment is editable before a single byte of PDF is written.

Questions about translating PDFs

Short answers, limits included

Do the tables and images survive?
Images and vector graphics are untouched: only text blocks get covered and rewritten. Table cells count as blocks of their own, so the grid stays put, though a much longer translation in a narrow cell shows up as an overflow.
Can I translate into Hindi, Russian or Chinese?
Not yet, and those languages are not offered rather than offered and broken. Writing them needs an embedded font file, which this tool does not carry. Offering them would hand you a PDF full of question marks.
What does an overflow mean?
That the translated block did not fit its original rectangle even at the smallest size allowed, so it was written anyway and may overlap what comes next. Every one is listed with its page. Edit that segment shorter and rebuild.
How does the glossary work?
One line per term: source, then target. It lands after the translation comes back, so it overrides the model. Leave the target empty and the term returns to its original form, which is how you protect a product name.
Does my document leave my computer?
Yes, and the panel says so before anything moves: the file goes to our server, the text goes to DeepSeek, our copy is deleted afterwards. The extraction and the rebuild both happen in your browser. Press cancel and nothing was sent.
Does the original text disappear from the file?
No. Each block is covered with a rectangle and the translation is written on top, so the source text is still inside the PDF and can be extracted or copied even though you cannot see it. That is what keeps the layout intact. If you are sending the document to someone and the original must not travel with it, choose the "Segments only" output and build the final document yourself, or run the result through the redaction tool, which rasterises the page so the text really is gone.

Updated on

Do more with your PDFs

Discover all the professional tools we have prepared for you. All in one place.