Translate Word documents to or from English while keeping your original layout and formatting
Built & maintained by Pappu Venkata Subbi Reddy, founder of Clacify · Updated July 2026
Document Translator turns a Word (.docx) document into another language while leaving everything except the text itself untouched — fonts, colors, styles, tables, images, headers, footers, and page layout all survive exactly as they were. It runs entirely in your browser using small, on-device AI translation models (one per language pair), not a metered cloud API, which means there is no character cap, no monthly limit, and no account to create — translate one document or a hundred, for free, forever. Each language pair's model downloads once (roughly 400-450MB) and is cached by your browser afterward, so every future translation with that same pair is instant with no further downloads. This matters for anyone who has ever tried to machine-translate a formatted document before: paste text into Google Translate and you get plain, reformatted text with no structure; upload to most "document translator" tools and you either hit a monthly character limit or the layout comes back broken. Because this tool edits the document's underlying XML directly rather than rebuilding it from scratch, the translated file opens in Word exactly where the original left off, just in a different language.
A .docx file is a zip archive of XML parts. The tool unzips it in your browser, walks every paragraph in the main document, tables, headers, and footers, and pulls out the plain text of each paragraph (concatenating its individual text runs, since a single sentence is often split across several runs by Word's own editing history). Each paragraph is sent as one unit to a small on-device translation model specific to your chosen language pair — running via ONNX in a background Web Worker so the page stays responsive — and the translated text is written back into the very same XML node the original text came from. Every formatting node (run properties, paragraph styles, table structure, embedded images) is left completely untouched throughout, because the tool only ever modifies text content, never the surrounding markup. The file is then re-zipped into a valid .docx. Paragraphs inside tracked-changes markup are skipped and flagged, since it's ambiguous whether pending edits should be translated before they're accepted or rejected. One side of every supported pair is English — that's the boundary of what real, verified translation models actually exist for.
Most free translation tools are a thin wrapper around a paid cloud API (Google Cloud Translation, DeepL, Azure Translator), all of which meter usage by character count — typically 500,000 free characters a month before billing kicks in. A tool built that way has to either cap what you can translate or eventually ask you to pay. This tool sidesteps that entirely by running a small AI model on your own device instead of a server: there's no per-character cost to manage because there's no API call being metered at all. The tradeoff is a modest one-time download per language pair (roughly 400-450MB) the first time you use that pair — after that, it's cached and every future translation with it is free and instant.
Document layout — margins, page breaks, paragraph styles, heading styles, tables, images, headers and footers — is preserved exactly, because none of that XML is ever rewritten. What isn't guaranteed is word-level formatting inside a single translated sentence: if three specific words were bold in the English original, machine translation can reorder and re-length the sentence in the target language, so there's no reliable way to know which translated words should inherit that same bold styling. This is a limitation of machine translation itself, not this tool specifically — it's the same simplification Google Docs' own built-in translator makes. Paragraph-level formatting (an entire heading, an entire bold paragraph) is unaffected either way.
Students applying to universities abroad translating transcripts and statements of purpose; professionals sending a resume or cover letter to an international recruiter; small businesses translating a contract, invoice template, or product spec for an overseas client; immigrants and families translating official letters and forms into their home language to share with relatives — anywhere a formatted Word document needs to cross a language barrier without losing the way it looks.
Translation quality is generally strongest for well-formed, grammatical source sentences — very short sentence fragments, or text that's really a label rather than a sentence (like a table header of just "Qty"), can translate less naturally than full sentences. If a document has tracked changes turned on, accept or reject them in Word first — the tool skips paragraphs with pending tracked changes rather than guessing which version to translate. Always use the built-in preview to sanity-check the result before relying on the downloaded file for anything formal.
Correct — because translation happens entirely inside your browser using a downloaded AI model, not a metered cloud API, there's no character cap, no daily limit, and no account needed. Translate as many documents as you like, for free, forever.
Yes, for layout: your fonts, colors, styles, tables, images, headers, and footers are left completely untouched — only the text content is translated in place. The one exception is word-level formatting inside a single translated sentence (e.g. exactly which 3 words stay bold) — because translation changes word order and length, that can shift slightly, the same way it does in Google Docs' own translator.
No. A small translation AI model downloads once per language pair to your browser (roughly 400-450MB, cached afterward) and every document you translate is processed locally on your device. Nothing is ever sent to a server.
English to/from Hindi, Spanish, French, German, Italian, Russian, Arabic, Chinese, Vietnamese, Dutch, Japanese, and Indonesian. One side of any translation must currently be English — direct translation between two non-English languages is not yet supported.