Rebuild research-paper PDFs as readable, one-column PDFs for the Supernote A5 X2 Manta. The output uses the Manta screen size, preserves equations and figures, removes page furniture, and leaves a right margin for handwritten notes.

The conversion pipeline is fully local after MinerU downloads its models:

PDF -> MinerU -> Markdown and assets -> Pandoc/XeLaTeX -> PDF

• macOS with the Palatino font
• Python 3.11, 3.12, or 3.13
• Pandoc
• XeLaTeX

On macOS, Pandoc and XeLaTeX can be installed with Homebrew:

brew install pandoc
brew install --cask mactex-no-gui

python3.13 -m venv .venv
. .venv/bin/activate
pip install -e .

MinerU may download several gigabytes of models on the first conversion.

pdf2pdf article.pdf article-manta.pdf

The default output has 8 mm page margins and a 30 mm right notes margin. Both are adjustable:

pdf2pdf article.pdf article-manta.pdf \
  --page-margin-mm 8 \
  --notes-margin-mm 30

The converter is also available as a Python function:

from pdf2pdf import convert_pdf

convert_pdf("article.pdf", "article-manta.pdf")

Pass an object with an extract(input_pdf, output_dir) method to convert_pdf(..., extractor=...) to use another local or API-backed extractor.

python -B tests/test_reflow.py

The test generates temporary synthetic PDFs and does not require bundled article files or OCR models.

OCR and formula recognition can make content mistakes. Very complex tables, multi-panel figures, and unusually long equations may require manual review.

This project was developed with assistance from OpenAI Codex. AI-generated changes were reviewed and tested before inclusion.