What does a PDF translator do?
A PDF translator takes a document written in one language and gives you what it says in another. This PDF Translator reads the text out of your PDF, sends that text to an AI model a few pages at a time, and puts each translated block right next to the original it came from. You drop the file, pick a language, press Translate, and read it page by page.
The PDF itself never leaves your device. Your browser opens the file and pulls the text out. Only that text goes to our AI provider. The file's bytes, its images and its metadata stay on your machine.
That's the opposite of how the category usually works. The typical flow is: upload the whole file to someone's server, make an account, hit a monthly file allowance, then get back a PDF you can't edit. Four steps, and three of them are in your way. Here you get the text, in blocks, next to the source. You can copy it, download it as a .txt file, or print it.
When to translate a PDF this way
Use it when you need to understand a document before you act on it. That's most of the time someone searches "translate pdf to english."
A lease or a contract in a new country. You moved to Lisbon, the landlord sent eight pages in Portuguese, and you sign tomorrow. You don't need a pretty copy. You need to know that the deposit is two months and that clause 7 says the rent can't go up in year one. Reading the original and the translation side by side lets you check clause 7 against clause 7.
A letter from a government office. Tax notices, residence permits, school forms. These are short, dense and full of dates and amounts. The layout doesn't matter. The deadline does.
A paper or a reading list for class. A 12-page article in German, a methods section in Japanese. You want the argument, not the typesetting. You can set From page and To page and translate only the section you're assigned.
A manual or a spec sheet. Numbered steps and tables of values. The translation keeps table rows as separate lines instead of melting them into a paragraph.
It works the same for a freelancer checking a client contract, a student with a deadline, or someone at a big company reading a supplier's terms on a side project. Same tool, same limits, same privacy.
How the PDF Translator works
The PDF Translator does three things, in this order.
1. Your browser reads the text. Each page's text comes out with its line breaks and paragraph gaps. A big vertical gap between lines becomes a paragraph break. That matters later, when a long page has to be cut.
2. The text is packed into runs. A run is one request to the AI. The model can write about 4,000 tokens in one answer, and a translation is often longer than its source. English into Hindi can take roughly twice the tokens. Japanese or Chinese into English can take roughly three times the words. A run sized by character count alone would get silently cut off halfway through a clause. A cut-off lease is worse than no translation at all.
So each run gets a weight that stands in for output length:
Weight = (Chinese, Japanese and Korean characters × 3) + (every other character × 1). Whole pages are packed into a run while its weight stays at or under 4,000. A page heavier than 4,000 is cut on its own, at the last paragraph break, else the last sentence end, else the last space.
"日本語です" is five characters and weighs 15. "Hello" is five characters and weighs 5. Russian, Arabic and Greek letters weigh 1, like Latin ones. Emoji count as one character each, not two.
3. Runs go out one at a time. Each finished block shows up as soon as it's back. If one run fails, that block gets a Retry button and the rest keep going. If a run comes back close to the model's limit, the block says This part may be cut off. and offers Retry in two halves.
One translation takes at most 10 runs. Here's what that means in pages. We ran each row through the tool's own planning engine:
| Text on each page | Pages per run | Most pages in one go |
|---|---|---|
| 900 characters (a sparse page, lots of headings) | 4 | 40 |
| 1,500 characters | 2 | 20 |
| 2,000 characters | 2 | 20 |
| 2,500 characters (a dense page of small print) | 1 | 10 |
| 1,000 Japanese characters (weight 2,890) | 1 | 10 |
| 1,500 Japanese characters (weight 4,334) | Each page splits into 2 runs | 5 |
If your range needs an 11th run, the tool tells you: These pages hold more text than one translation takes. Pick a smaller page range. Translate pages 1–20, then 21–40. Nothing about the first batch is lost when you start the second.
A worked example: an 8-page lease
Say you drop a lease into the PDF Translator. It has 8 pages of about 1,500 characters each. That's a page of contract text with some white space around the clauses.
- Drop
lease.pdf. The strip reads 8 pages · 4 runs. - Leave Translate from on Detect automatically. Translate into already says English on the English site. It defaults to your page's language, so the Brazilian Portuguese site starts on Portuguese (Brazil).
- Press Translate. Four blocks come back, labelled Pages 1–2, Pages 3–4, Pages 5–6 and Pages 7–8. Each run weighs 3,000, comfortably under the cap.
- Download the result. The file is named
lease-en.txt, with each block's label on its own line above its text.
Each run costs one AI credit. Without an account you get 3 credits, which covers the first six pages of this lease. After the third run the page says Translated pages 1–6. You're out of AI runs for now. The three finished blocks stay put. Sign up, press Continue, and it sends only the fourth run. A new account comes with 15 credits. Members get 1,500 a month.
Translating the same text into the same language again within 7 days costs nothing. Finished translations are cached by the text itself, not by who you are.
How to check a machine translation before you rely on it
The model is told to translate meaning rather than word for word. It keeps names, URLs and email addresses as they are, and writes numbers the way the target language writes them. So 1.250,00 in a Portuguese lease should come back as 1,250.00 in English. That's good enough to understand a document. It isn't a sworn translation. Having the original right next to it is what makes checking fast.
Spend your attention where mistakes cost money:
- Numbers and amounts. Match every figure in the translation to the original. Decimal commas and thousands separators flip between languages.
- Dates. 03/04 is March 4 in one country and April 3 in another. Find the date in the source and read it yourself.
- Negations. "May not", "shall not", "unless". One dropped word flips a clause.
- Defined terms. If the contract defines "the Premises", the same word should mean the same thing in every block.
- Tables. Rows come back as lines. Check each value still sits next to its label.
For anything with legal weight, use the translation to find the clauses that matter. Then have those clauses checked by someone who reads both languages. The tool gets you to the right three paragraphs in a minute instead of an evening.
Two settings help the model. If a PDF mixes languages, like an English contract with a Spanish annex, set Translate from by hand instead of leaving it on Detect. And translate the section you need rather than the whole file. A tighter range means fewer runs and fewer credits.
Limits, and what to do when a PDF won't translate
If a file isn't working, the fix is almost always one of these. Start at the top.
"This PDF has no text to read"
It's a scan: a photo of pages with no text behind it. Phones, office scanners and fax services make these. Run it through OCR PDF first, which adds a text layer, then bring the new file back.
"This PDF is password-protected"
The browser can't read an encrypted file. Open it with Unlock PDF, then translate the unlocked copy.
"That PDF is over 100 MB"
Big files are usually big because of images, which don't get translated anyway. Cut out the pages you need with Split PDF.
"This part may be cut off"
The run came back close to the model's output limit, so the end of it may be missing. Press Retry in two halves. The tool re-plans that block at half the budget, so every piece has room to finish.
The layout is gone
That's by design. You get the text, page by page, not a copy of the original layout. Translated text runs longer or shorter than the source, so it doesn't fit the boxes it came from. Arabic runs right to left. Chinese and Hindi need fonts the original never embedded. A PDF translator that promises the same layout tends to shrink, cram or overflow the text. If you want a PDF anyway, Print or save as PDF prints the translation only, and your browser draws every script correctly.
Related PDF tools
Want the original text without translating it? PDF Extract Text pulls it out as plain text, in your browser too. Need to edit the source document instead of reading it? PDF to Word turns it into a .docx. Got your translation as a .txt and want it back as a clean document? TXT to PDF does that in one step.
Frequently asked questions
Can I translate a PDF to English from any language?
Leave Translate from on Detect automatically and the model works out the source language itself. That covers most documents. You can translate into 39 languages, including English, Spanish, both Brazilian and European Portuguese, German, Polish, Ukrainian, Turkish, Hindi, Bengali, Vietnamese, Indonesian, Swahili, and Chinese in both Simplified and Traditional.
Can I translate a PDF into Arabic or Hebrew?
Yes. When the target is Arabic, Hebrew, Persian or Urdu, the translated blocks read right to left, on screen and in the print view. The original blocks set their own direction from their text, so an English source still reads left to right beside its Arabic translation.
Does it keep headings and tables?
It keeps the structure it can see. Paragraph breaks stay, headings stay on their own lines, and table rows stay as separate lines instead of being merged into a paragraph. Lines broken by the PDF's column width get joined back up. Words hyphenated across a line break get rejoined. What stays behind are pictures, colours and positions on the page.
If I translate the same page twice, does it cost twice?
No, for 7 days. A finished translation is cached against the exact text and the target language. Send the same page into the same language again and it comes straight back, with no credit spent. Switch the target language, or change a single word of the source, and it counts as a new run.
What happens if one part fails?
Only that block is marked, with That part didn't translate. That's on us. and a Retry button. Every other block carries on and stays on the page. If the AI provider hits its monthly limit on our side, you'll see that instead, and your translated pages are still there to copy or download.
Can I get the translation as an editable document?
Not as a .docx. You get three ways out: Copy translation, Download .txt, and Print or save as PDF. Paste the copied text into any document editor and it arrives as plain paragraphs, with each block's page label kept as a line of its own.
Where does the money go?
The AI runs cost us real money, which is why they use credits. Membership pays for them without ads. And 10% of every dollar Microapp earns goes to charity, off the top, audited quarterly.