How to Translate a PDF Online: Free and AI Options
To translate a PDF, open swiftin.dev/translate-document, drop the file, pick a language, and hit Translate. The free path uses Google or Bing with no account at all; the AI path understands context and tone. Either way the file itself never leaves your browser, and the translated PDF keeps the original layout.
The fastest way: translate a PDF for free
Three steps, no sign-up:
- Open the document translator in any browser.
- Drop your PDF (up to 25 MB). The preview renders every page right in the tab.
- Pick the target language, choose Google or Bing, and press Translate. When the preview settles, hit Download.
That is the whole flow. The free engines run browser-direct, so there is no account wall, no email capture, and no "3 pages left today" counter. Word, Excel, PowerPoint, EPUB and subtitle files go through the same three steps.
Free engines or AI: which should you pick?
Google and Bing are fine for getting the gist of a manual or an invoice. AI translation earns its keep when the text has voice: contracts where one mistranslated clause matters, books where tone carries the page, anything with idioms, slang, or domain terms. AI reads whole blocks in context instead of sentence by sentence, so terminology stays consistent across the document.
| Feature | Free engines | AI engines |
|---|---|---|
| Engines | Google, Bing | Pick your AI model, or bring your own key |
| Account needed | None | Extension signed in (or just the extension, with your own key) |
| Cost | Free, unlimited | AI tokens, 50,000 free every month |
| Context awareness | Sentence-level | Block-level, consistent terminology |
| Formats | All, including Excel and PowerPoint | PDF, Word, EPUB, text, Markdown, HTML, subtitles |
The AI path runs through the Swiftin browser extension, which translates the rendered preview in place, the same engine that translates live webpages. The free plan includes 50,000 AI tokens a month; when they run out, Google and Bing keep working, unlimited.
Does the translation keep the PDF layout?
Yes, and this is the part most online converters get wrong. In replace mode the original page stays exactly as it is: colours, images, tables, stamps, signature lines. Each text block is covered with a patch sampled from its real background colour, and the translation is drawn on top in the same position, auto-shrunk to fit the box when the target language runs longer.
Prefer to keep the original text visible? Bilingual mode copies every original page unchanged and inserts a clean translation page right after it, useful when you need to check the translation against the source, or hand both to someone else.
And if the machine gets a block wrong, the preview is an editor: click a block to rewrite it, drag it, resize it, change the font size, or delete it entirely. Your edits are merged into the downloaded PDF. Non-Latin scripts are handled too, the assembler embeds a matching font for Cyrillic, CJK, Arabic, Hebrew, Devanagari and Thai.
What about Word, Excel, PowerPoint, EPUB and subtitles?
The same page translates fourteen formats: docx, xlsx, pptx, epub, txt, Markdown, HTML, and the whole subtitle family (SRT, VTT, ASS, SSA, SBV, LRC). Two notes worth knowing:
- Bilingual downloads are available for text, Markdown, HTML, EPUB and Word: the file comes back with the original and the translation stacked together, which readers of bilingual page mode will recognise.
- Excel and PowerPoint run on the free engines only, the cell and slide structure has no clean way to preview-translate with AI, so those two formats always take the free path.
Is it safe for contracts and private documents?
The file itself never leaves your browser: parsing and text extraction happen locally, and only the extracted text is sent to the translation engine. On the free path that engine is Google or Bing, and their policies apply to the text they process. On the AI path the text goes through Swiftin's backend, where nothing is stored by default, there is no server-side history unless a Team owner explicitly opts in.
For NDA-bound or regulated documents, that boundary matters: no third-party file storage, no document sitting in an upload bucket, no retention you have to chase down later.
When this is the wrong tool
Honest scope, three cases:
- Scanned PDFs. If the PDF is a photo of a page with no text layer, there is nothing to extract and nothing to translate. You need OCR first; Swiftin does not do OCR.
- Files over 25 MB. The parser runs in your browser, and very large files are capped to keep the tab responsive.
- Pixel-perfect print production. Replace mode preserves layout remarkably well, but if you need typeset-quality output for print, plan on a manual pass in a desktop publishing tool afterwards.
For everything else, the browser is the shortest path from "PDF in a language you do not read" to "same PDF, in yours".
FAQ
Keep reading
Private Translator Extension: What "No History" Actually Means
Most browser translators retain everything you translate on their servers. Swiftin does not store translations by default. What that means for sensitive text.
Bilingual Reading Mode: Side-by-Side vs Replace-in-Place
Two reading modes, one decision: see the original AND the translation at the same time, or replace the original cleanly. This guide explains when each one wins.
Why Free Translator Extensions Quit on You (And How to Avoid It)
Free translator extensions quit in four ways: daily caps, API-key walls, sudden paywalls. Here is how a fallback layer keeps translation running after the quota.