PDF Translator (English ↔ Korean, Spanish, Portuguese, French)

Upload a text-based PDF and get a paragraph-by-paragraph translation: English ↔ Korean, English ↔ Spanish, English ↔ Portuguese, English ↔ French, or English → German. Review original and translated text side by side (and edit it), then download as plain text, HTML, or a PDF that keeps the original paragraph positions. Nothing is uploaded to a server.

Loading the tool...

When to use it

Read an English paper or manual in Korean

Upload an English PDF with real text and leave the direction on “Auto-detect”. Press “Start translating” and it goes paragraph by paragraph; when it is done you can read the original and the translation side by side.

Fix the draft before you hand it in

Machine translation is a draft. Edit awkward sentences in “Translation (editable)”, then save with “Download as .txt” or “Download as HTML”.

Get a translated PDF that keeps the page look

Once you are happy with the text, press “Download format-preserving PDF” to get the translation placed where the original paragraphs were. Tables and multi-column pages can shift a little.

Translate a long document in sections

Long files default to the first 5 pages. Change “From page” and “To page” to work in ranges, and use “Pause” and “Resume” to manage the time.

How to use it

  1. Drop in the PDF you want translated, or click to choose it. Only text-based PDFs are supported; for a scanned image PDF, first add a text layer with OCR PDF.
  2. Pick a translation direction (English→Korean / Korean→English) or leave it on auto-detect. For Korean→English you can choose between the fast model (~80MB) or the accuracy-first multilingual model (~630MB). For long documents you can choose a page range (default: the first 5 pages).
  3. Click "Start translating". The first run downloads a translation AI model once (about 80MB for the fast Korean→English model, or about 630MB for the multilingual model), then translates paragraph by paragraph, showing progress and an ETA. You can pause or cancel at any time.
  4. Once done, review the original and translation side by side and edit the translation directly. When you're happy with it, download as plain text (.txt), HTML, or a PDF that keeps the original layout.

How it works

pdf.js extracts text with positions, paragraphs are rebuilt by our code, and translation runs in your browser with transformers.js and ONNX Runtime. The translation model is downloaded once from our model storage (models.dagotools.com, with Hugging Face as a fallback): about 80 MB for the fast Korean-to-English model (opus-mt-ko-en) or about 630 MB for the multilingual one (M2M100). Your PDF and its text are never uploaded, and scanned image-only PDFs can't be translated. Your file is processed in this browser and is not uploaded. To check this yourself, watch the browser's DevTools Network tab while you use the tool (step-by-step guide). Also available as a free REST API for developers and AI agents (edge-processed, zero retention).

How paragraphs are detected

Dagochim's PDF Translator reads every text fragment inside the PDF (its position, font size, and line spacing) and automatically groups lines that look like the same paragraph. Lines with a noticeably larger font size are treated as headings and shown in bold; lone numbers sitting alone near the top or bottom of a page (likely page numbers) are excluded from translation. This grouping isn't perfect though - documents with tables or multiple columns may have paragraphs grouped slightly differently or out of the expected order.

How the format-preserving PDF is built

When you download the "format-preserving PDF", the original PDF pages are kept as-is: the tool draws a white rectangle over each original paragraph's area, then draws the translated text back into that same spot. If the translation is longer than the original, the font size automatically shrinks to try to fit the same box. To display both Korean and Latin characters, the tool embeds (and subsets) the Nanum Gothic font inside the PDF, which can slightly increase file size. Tables, multi-column layouts, and text overlaid on images may end up slightly misaligned - please keep that in mind for complex documents.

About the translation AI model

This tool lets you choose between two open-source translation AI models. For Korean→English, it defaults to Helsinki-NLP's opus-mt-ko-en (about 80MB) for much faster translation; for every other direction (or if you pick the "accuracy-first" option), it uses M2M100, the multilingual model released by Meta AI (about 630MB). Both models run directly in your browser and aren't downloaded automatically - they only download once you click "Start translating". We recommend using Wi-Fi for that first download; after that, the model is cached in your browser and future translations start instantly. Paragraphs you've already translated are also cached, so re-running a translation on the same text is instant.

Open-source credits

More tips

1. First, check whether your PDF actually contains text

Before you try to translate anything, it helps to understand what kind of PDF you're dealing with. Most PDFs created from Word, Google Docs, Hangul (HWP), or a web page retain real text data internally - the letters, their positions, and their font sizes are all stored as structured data, even though visually it just looks like a printed page. These are called "text PDFs," and they can be processed directly: a computer program can read every character, its coordinates, and its size without needing to guess anything.

On the other hand, a PDF made by scanning a paper document with a scanner or photographing it with a phone camera is fundamentally different. Even though it looks exactly like a normal page of text, what's actually stored inside the file is a picture - a grid of pixels - with no character information attached at all. A computer has no way to know that a particular cluster of pixels represents the word "Agreement" unless it runs optical character recognition (OCR) first. The quickest way to tell the difference yourself: open the PDF in any viewer and try to click-and-drag to select some of the text. If a highlighted selection box appears around the words, it's a text PDF. If nothing gets selected no matter where you drag, it's a scanned image PDF.

2. If it's a text PDF: use Dagochim's PDF Translator directly

If your document is a text PDF, you can go straight to Dagochim's PDF Translator and drop the file in, or click to select it. Once it's loaded, the tool samples a few paragraphs and automatically guesses whether the document is mostly English or mostly Korean, suggesting a translation direction accordingly; you can also override this and pick "English → Korean" or "Korean → English" manually if the auto-detection gets it wrong (which can happen with documents that mix both languages heavily, like a bilingual contract). For longer documents, you can choose a specific page range to translate instead of the whole file - this defaults to the first five pages, which is usually enough to preview the quality before committing to a full run. If you do need the entire document translated, you can widen the range, but keep in mind that translation time scales with the number of paragraphs, so a 40-page report will take noticeably longer than a two-page memo.

3. If it's a scanned PDF: run OCR first

If you try to load a scanned PDF into the translator, it will detect that there's essentially no extractable text and show a message pointing you toward OCR instead of silently producing an empty or broken result. In that case, the recommended workflow is to first open Dagochim's Image to Text (OCR) tool, which reads the printed characters out of each page image and turns them into actual selectable, copyable text. OCR does a solid job with clearly printed text at reasonable resolution, but it can struggle with handwriting, low-quality or skewed scans, unusual fonts, or very small print - so it's worth skimming the OCR output for obvious errors (a misread character here and there, merged words, or garbled lines) before copying it over for translation. Once you have clean extracted text, use it as your working document; this tool itself translates text-based PDFs and has no box for pasting text.

4. Know what to expect: a one-time model download (about 630MB)

It's worth understanding how this translator actually works under the hood, because it affects what to expect the first time you use it. Rather than sending your document's text to a remote server to be translated and sending the result back (which is how most "free online translator" websites operate), this tool downloads an open-source translation model (Opus-MT for Korean to English by default, or Meta AI's multilingual M2M100 for other directions and the accuracy-first option) and runs it entirely inside your own browser using WebAssembly. That means the actual translation computation happens on your device, not on Dagochim's servers or anyone else's.

The tradeoff is that model files are not small - expect roughly 630MB to download the first time you click "Start translating" for English to Korean and most other directions (Korean to English uses a smaller model of about 80MB by default). After the first download, the model is usually kept in the browser's own storage, so later translations start much faster without downloading it again. If the browser clears that storage (for example when you clear site data or the device runs low on space), the model downloads again. If you're on a mobile data plan or a slow connection, it's worth running that first translation while connected to Wi-Fi so the download doesn't eat into your data allowance or take an unexpectedly long time.

Frequently asked questions

Can it translate scanned PDFs (images)?

This tool only translates text-based PDFs directly. For a PDF made from scanned pages or photos, first turn it into a PDF with real text using Dagochim's OCR PDF tool, then upload that PDF here again to translate it.

How good is the translation quality? Is it as accurate as a professional translator?

This produces a machine-translation draft using open-source AI models (opus-mt-ko-en for Korean→English, M2M100 for every other direction). It handles everyday sentences reasonably well, but technical terminology, complex sentences, and formal documents like contracts can come out wrong or awkward. Always have a human review important documents after translation.

Is my file sent to a server?

No. Reading the PDF, extracting text, AI translation, and generating the new PDF all happen locally in your browser. The first time you click translate, the tool downloads a translation AI model once (about 80MB for the fast Korean→English model, or about 630MB for the multilingual model) - that download is just the model itself; the PDF or text you work with is never uploaded anywhere.

Does the format-preserving PDF look exactly like the original?

It stays close but isn't pixel-perfect. The tool covers each original paragraph's area with a white box and draws the translated text back into that same spot, so simple single-column documents usually look clean, but tables, multi-column layouts, or text overlaid on images may shift slightly or auto-shrink in font size.

Does it translate languages other than English and Korean?

Yes, with English on one side. Besides “English → Korean” and “Korean → English” you can pick English ↔ Spanish, English ↔ Portuguese, English ↔ French and English → German. German → English and Korean ↔ Spanish/Portuguese are not offered because test translations had too many mistakes. “Auto-detect” only tells English and Korean apart, and languages such as Japanese or Chinese are not supported.

Should I pick the fast or the accuracy-first model?

The choice appears for Korean to English only. The fast model is about 80 MB and light to start; the accuracy-first multilingual model is about 630 MB and takes longer to download. English to Korean and every Spanish, Portuguese, French and German direction use the multilingual model.

Is the model downloaded every time?

Only the first time you press “Start translating”; it is then cached in your browser. Because the files are large, Wi-Fi is recommended for the first run. Paragraphs translated before are reused instantly.

What if translation stops partway?

Use “Pause”, “Resume” or “Cancel”, and “Retry” if something fails. If the browser cannot run the model, try a recent Chrome or Edge.

Can I edit the translation?

Yes. In the "Translation (editable)" view you can change sentences directly, and the files you download afterwards include your changes. Compare names and numbers with the original once more.

Which formats can I download?

"Download as .txt", "Download as HTML" and "Download format-preserving PDF". Plain text is handy for copy and paste. Choose the format-preserving PDF if you want the closest look to the original.

Tools you might also like

Related guide

See How to translate an English PDF to Korean and keep the formatting for a step-by-step walkthrough.