About this tool
The AI Document Translator turns a Khmer or mixed-language document into a translated PDF that keeps the original layout. Upload a PDF, a Word file (.docx), a plain text file, or a photo of a page — PNG, JPEG, or WebP — up to 100 MB, choose a target language, and watch the job move through splitting, extraction, translation, and rendering stages with live progress on the page. When it finishes you download a new PDF where every translated line is placed back where it came from, so tables, letterheads, and signatures stay in place. The tool is built for Cambodian use cases first: official letters, contracts, certificates, invoices, and school or government paperwork that mix Khmer and English. The free output carries a small InKhmer mark in the page footer. If you need a clean copy of a single document — for a visa file, a court submission, or a client deliverable — you can remove the mark for that one document with a one-off Bakong KHQR payment. No account is required and there is no subscription. Your original file is stored on our server for 24 hours so the pipeline can read it, then deleted; the translated PDF is kept for 7 days so you can re-download it.
How it works
When you upload a file, the browser sends it as multipart form data to the InKhmer API, which streams it straight to object storage without buffering the whole body — a byte counter aborts the upload immediately if it crosses the 100 MB cap. The server then creates a job row and publishes it to a shared translation queue. A worker drives the job through a fixed state machine — queued, splitting, extracting, translating, rendering, then done — and every transition is recorded so the page can show live progress over a server-sent events stream. PDFs are split into pages and each page is sent to a vision model that returns structured JSON describing the text spans and where they sit on the page. A second pass translates each span into your chosen target language, preserving the span positions so the renderer can rebuild a PDF where each translated line lands in the same place as the original. Pages the model cannot read with enough confidence are retried with a stronger model; if they still fail they are kept in the original language and the job finishes as partial. The rendered PDF gets an InKhmer footer mark on every page unless the watermark has been removed for that document.
Use cases
- Translate a Khmer employment contract or labor ministry form to English for an overseas employer or embassy appointment
- Turn a family record — birth certificate, marriage certificate, family book — into English or French for a visa application file
- Read a supplier quotation or purchase order written in Chinese or Thai by translating it into Khmer or English
- Translate school transcripts and diplomas for foreign university admissions
- Convert a scanned Khmer letter or notice photo into a translated PDF that keeps the official letterhead layout
- Translate terms of service, lease agreements, or banking documents into Khmer to review before signing
- Prepare a bilingual deliverable with the side-by-side variant so the original and translation print on alternating pages
Frequently asked questions
What file types and sizes can I upload?
PDF, Word (.docx), plain text (.txt), and page photos in PNG, JPEG, or WebP format, up to 100 MB per file. The server checks the file type and streams the upload with a hard byte cap, so oversized or unsupported files are rejected with an error before any translation job starts.
Which languages can I translate into?
The target-language picker currently offers English, Thai, Vietnamese, Chinese (Simplified), French, Japanese, Korean, and Spanish. The source language is detected automatically — Khmer and mixed-language documents work best, but other source languages are accepted. More target languages can be added as the model quality for them is verified.
Is the original layout really preserved?
Yes. The translation step returns every text span with its original position, and the rebuilt PDF places each translated line where the source line sat. Very long translations can overflow a tight cell or narrow column, so give the result a quick read before you submit it somewhere official.
What does the free version include, and what is the InKhmer mark?
The free output is a complete translated PDF; every page footer carries a small “Translated by InKhmer” mark. To remove the mark on a single document you pay a one-off fee via Bakong KHQR and re-download a clean copy — no account or subscription.
Where is my file stored and for how long?
Your original file is uploaded to the InKhmer server and stored in object storage for 24 hours so the pipeline can read it, then deleted. The translated PDF is kept for 7 days so you can download it again, then deleted too. The AI that reads the text runs entirely on our servers.
What happens if a page cannot be read?
Pages that fail extraction or translate with low confidence are retried once with a stronger model. If they still fail, those pages are kept in the original language and the job finishes with a “partial” status — the result page lists them so you can jump straight to the pages that need a manual pass.
Why is my translation taking a while?
Translation runs page by page, so long or dense documents take proportionally longer. The progress bar shows which stage the job is in and how many pages are done. If the job fails outright you get a bilingual error message instead of a silent timeout.
Can I translate a whole book or very long PDF?
The 100 MB cap is on file size, not page count, but very long documents are slow and a single low-confidence page can mark the whole job partial. For book-length work, splitting the PDF into chapters before uploading gives faster, more reliable results.
Is the translation accurate enough for official use?
The models handle formal Khmer and common mixed-language layouts well, but machine translation is not certified translation. Embassies, courts, and ministries may require a human-certified translation — treat the output as a faithful draft you can verify or hand to a translator to finalize.
Data and accuracy
Translation is produced by server-side AI models hosted through the InKhmer translation service — no model key or prompt ever reaches your browser. The pipeline reads each page, translates its text spans, and rebuilds a PDF that puts every translated line back where it came from. Files and outputs are retained only as described above.
Sources
- Bakong — Cambodia’s national KHQR payment system — National Bank of Cambodia
- Unicode Khmer script code chart — Unicode Consortium
Content last reviewed 10 October 2026