AI PDF Translator: Keep Layout, Translate Everything - Open Source
An AI PDF translator that keeps the layout: MinerU or Nougat extract the text, an LLM translates, and you rebuild the document.
💡 What You Will Learn
An AI PDF translator that keeps the layout: MinerU or Nougat extract the text, an LLM translates, and you rebuild the document.
Translating a PDF used to mean copy-pasting text into Google Translate and losing every table. The modern open-source route extracts structure first, translates second, and rebuilds the document with layout intact.
The Pipeline
MinerU (77,061 stars) is the current workhorse: it converts PDFs (including scanned ones) into clean markdown with tables and headings preserved. Nougat (10,057 stars) is the academic-document specialist. The markdown goes to an LLM - DeepSeek-V3 (104,114 stars) for quality, or a local model via Ollama (177,874 stars) for privacy. Rebuild the translated markdown into PDF with your normal toolchain.
Practical Notes
Scanned PDFs need OCR - MinerU handles it, but expect 1-2 minutes per page on CPU. Math formulas are the weak spot: Nougat was built for them, general models will mangle them. For contracts and specs, always have a human review numbers and legal terms.
FAQ
Q: Is it free?
A: Yes - all tools are open source; the only cost is compute, and local models make even that near-zero.
Q: Can it handle 100-page documents?
A: Yes, in chunks - process 10-20 pages at a time and stitch the translated markdown together.
