Folia Tools
All tools › Read & analyze › Translate PDF
Translate PDF
A clean translated copy, page by page. Free monthly allowance — needs an account.
or drop it here
Translate PDF reads the PDF in your browser and sends only the text it finds to Folia Cloud. It asks before anything leaves.
How to translate PDF
Translation reads the text of your document and writes it out in the language you choose, laid out on clean pages.
- Select your PDF. Any text-based document. A scan needs OCR first.
- Choose the language. Detect the source automatically, or set it yourself.
- Download. A translated PDF, plus a plain-text copy you can paste anywhere.
Text goes up; a freshly typeset document comes back
Your browser reads the text out of the PDF in reading order and sends only that. What returns is set on clean A4 pages at 11 pt, with a font chosen per script — the Korean, Japanese, Chinese, Arabic, Thai and Devanagari faces are embedded in the file itself, and right-to-left text is reordered properly rather than reversed. Lines are re-flowed into paragraphs rather than kept where they were, and a line that comes back with a bullet or a number in front of it is laid out as a list item. Images, logos, diagrams, table rules and column structure are not part of that text stream and do not come across.
So a two-column paper comes back as one column, and a table comes back as its cell text in reading order. That is why a .txt file arrives alongside the PDF: when the original layout matters, the plain text is what you paste back into the Word file the PDF was made from. Fonts are embedded whole rather than subset, so a Korean, Japanese or Chinese translation carries a few megabytes of font whatever its length.
Every section is translated on its own
The text is cut into sections of about 3,000 tokens at paragraph boundaries, and each is sent as its own request; one job caps at 120 sections, and a longer document should go through Split PDF first. Working one section at a time is what stops a dropped connection or a sleeping laptop from costing you the whole document: the finished sections come back with a note saying where it stopped, and only those were charged.
The price is that no section sees the others. Translating into Korean pins one formal register across the whole job, because mixed speech levels are the first thing a reader notices — but a company-specific term or an unusual proper noun can still come out two ways across a long file. Numbers, dates, URLs, code and proper nouns are meant to pass through unchanged unless they have a standard translated form, so a spot check on those is quicker and more useful than a full re-read.
Questions
Why does this need Folia Cloud?
Translation runs a large language model, which cannot fit in a browser tab. It is also the most expensive tool to run — the output is as long as the input — so it comes with a subscription rather than being ad-supported.
Does the translation keep the original layout?
No, and no tool honestly can. Translated text is a different length from the original — Korean runs to roughly 43% of the characters of the same English text, and translating the other way expands it — so lines and page breaks have to move. Forcing the original geometry would overlap or clip the text. Folia lays the translation out properly instead, with the correct fonts for every script, so the result reads cleanly.
What happens to my file?
Only the text is sent — not the file itself, and not any images. It is processed and deleted as soon as the job finishes, is never stored, and is never used to train anything. Every other tool on this site uploads nothing at all.
Which languages can it translate between?
Fifteen: English, Korean, Japanese, Chinese, Spanish, French, German, Portuguese, Italian, Russian, Arabic, Hindi, Vietnamese, Indonesian and Thai. The source language can be detected for you or set by hand, and any pair among them works — each section is sent as a single instruction naming the two, with no staging language in between.
Will Korean, Arabic or Thai render properly, or come out as empty boxes?
They render. The face for each script is embedded in the PDF, so the file opens correctly on a machine that has none of those fonts installed, and Arabic is laid out right to left. One weight per script is bundled, so bold in a non-Latin script is drawn as a double strike rather than a true bold face.
Will the translation have the same number of pages as the original?
Rarely. The text is re-flowed onto fresh pages and a translation is a different length from its source, so the same report comes out shorter in Korean and longer in German or Thai. Page-for-page correspondence is not something the file can keep — if you need to point someone at a passage, cite the heading rather than the page number.
Are my files uploaded to a server?
The file itself is not. This is one of only two tools on the site that send anything: your browser reads the PDF on your own machine, extracts the text, and sends that text — never the file, never its images — to Folia Cloud, because a model this size cannot run in a browser. It is processed, returned, and deleted when the job finishes. It is never stored and never used to train anything. The other 38 tools upload nothing at all.
Is there a file size limit?
Reading the PDF happens on your device, so that part is bounded by your own memory, and we show the number on the page. Beyond that a single job is capped at a few hundred pages of text so it cannot run away with your Cloud AI credits. Longer documents are best split first.
Is it really free? Do I need an account?
Not this one. This tool runs on Folia Cloud, which costs real money per document, so it needs a signed-in paid plan and spends Cloud AI credits from it. Every other tool on this site is free, unlimited and needs no account.