Scanned PDF to EPUB with Mistral OCR

Uses Mistral OCR to convert a scanned PDF to markdown and then to an epub. The main motivation is to be able to read scanned pdfs on my e-reader. There is also an optional automatic translation feature, but the accuracy may vary.

Install

1. Install pandoc from https://github.com/jgm/pandoc/releases/
   • check installation with pandoc --version
2. Make python 3.14 virtual env and install requirements

   python -m venv .venv
   source .venv/bin/activate
   pip install -r requirements.txt

3. Set MISTRAL_API_KEY environment variable to your Mistral API key.

Usage

Converting a PDF to EPUB.

python main.py path/to/input.pdf path/to/output.epub

Optional: translate to target language (e.g., French):

python main.py path/to/input.pdf path/to/output.epub --translate_to french

The language parameter is flexible, e.g., fr, fr-FR, french, etc. should all work.

API Costs

Costs depend on the document. In my experiments it is about 20 cents per 100 pages for OCR, and about 3-5 cents per 100 pages for translation.

Approach

1. Mistral OCR to get markdown text from each page, including images.
2. Mistral OCR Annotation on the first 10 pages of document for getting book metadata following pandoc's epub metadata.
3. Optional: use mistral-small-latest to translate the markdown in chunks (concurrently for speed).
4. pandoc to convert markdown to epub