pdf-craft: Convert Scanned PDFs with DeepSeek OCR

Overview

PDF Craft is a Python library and online service for converting scanned books, research papers, and technical documents into editable Markdown or reader-friendly EPUB. It is built around DeepSeek OCR from DeepSeek, with support for DeepSeek OCR 2 and Baidu's Unlimited OCR as well. The online service lets readers convert files in a browser; the Python library is for developers who want to integrate conversion or run OCR themselves.

The project targets scans whose text and structure need to be recovered for editing, searching, or reading in an ebook app. It can also translate PDF and EPUB content, create bilingual EPUBs, and write translations onto the original PDF pages.

Key Features

• Convert scanned PDFs to Markdown with text and image assets.
• Create EPUBs with book metadata and a table of contents.
• Extract document structure including chapters, footnotes, tables, formulas, and images.
• Use DeepSeek OCR, DeepSeek OCR 2, or Unlimited OCR with remote or local configurations.
• Translate during conversion or translate existing EPUBs, with replacement or bilingual output modes.
• Write translated text back onto the source PDF pages.
• Save extraction results in .pcex files for reuse in later processing.
• Choose browser-based conversion or integrate conversion through synchronous and asynchronous Python APIs.

Use Cases

• A reader wants to turn a scanned book into an EPUB without installing Python or configuring OCR, and can use PDF Craft Online.
• A researcher needs scanned papers converted to searchable, editable Markdown, with image assets retained alongside the text.
• A developer is building a document workflow and wants Python APIs plus reusable extraction files for later rendering or translation.
• A reader or publisher wants a translated or bilingual EPUB, or a translated PDF that preserves the original page layout as a starting point for review.

Getting Started

For the Python library, install it with:

python -m pip install pdf-craft

The quick-start example uses a compatible remote OCR service. See the README and installation guide for setup details. To convert without installing Python or configuring OCR, use PDF Craft Online.

Alternatives

• marker: Marker converts PDFs into Markdown and other structured formats, but does not focus on EPUB creation or document translation.
• gptpdf: gptpdf uses vision models to convert PDF pages to Markdown, while pdf-craft centers on DeepSeek OCR and also offers EPUB and translation.
• docling: Docling handles PDFs and many other formats for structured document workflows, rather than focusing on scanned books, EPUBs, and translation.
• indra: Indra converts PDFs and images to Markdown, text, or JSON with CPU-based OCR, but does not offer EPUB conversion or translation.

Considerations

• The project is active, with regular commits and releases, but its contributor base is small and work is highly concentrated in one contributor. Recent pull requests have been merged regularly, while the available issue-handling data shows little recent issue closure.
• The package requires Python 3.11 through 3.13 and Poppler. Remote OCR needs a compatible service URL and credentials; local OCR additionally requires CUDA-enabled PyTorch, sufficient VRAM, and model files.
• Translation and some metadata or table-of-contents analysis can involve separate remote LLM services. Review what content is sent to configured services before processing sensitive documents.
• Conversion quality depends on scan quality, page layout, and OCR model. The README recommends checking a representative document before processing a larger collection.
• Writing translations onto source PDF pages requires Ghostscript and suitable local fonts, and the resulting layout should be checked against the original.
• The project is MIT-licensed, but third-party dependencies and selected OCR models retain their own licenses.