pdf-craft

Convert scanned PDFs to Markdown and EPUB, with optional translation and translated PDF output.

From scanned pages to usable documents

PDF Craft is a Python library for scanned books and academic or technical documents. It extracts page content and organizes body text, chapters, tables of contents, footnotes, tables, formulas, and images for further editing and reading.

Markdown: edit, search, and process the content.

EPUB: read the book in an ebook reader.

Results depend on scan quality, page layout, and the OCR model. Check a representative document before processing a larger collection.

When you need EPUB bibliographic metadata, opt in to front-page OCR metadata extraction with a separate metadata LLM. PDF Craft verifies every accepted value against OCR evidence and uses PDF file properties only to fill missing fields, never to override printed book information.

What you can do

Choose how to use it

Remote OCR sends pages to the configured service and does not require local CUDA. Local OCR runs on your machine; using a remote LLM for translation or table-of-contents analysis still sends the corresponding content to that service.

Preview the online app

Quick start

This example uses remote OCR to convert a PDF to Markdown. Prepare Python 3.11–3.13, Poppler, and a working DeepSeek OCR-compatible service configuration. See the installation guide for Poppler setup.

1. Install

python -m pip install pdf-craft

2. Convert a PDF

Place input.pdf in the directory where you run your script. Replace the URL, API key, and model name with your service configuration:

from pdf_craft import DeepSeekOCRVendorConfig, PDFCraft, PDFOptions

craft = PDFCraft(
    pdf=PDFOptions(
        ocr=DeepSeekOCRVendorConfig(
            base_url="https://example.com/v1",
            api_key="your-api-key",
            model="deepseek-ocr",
        ),
    ),
)

craft.convert_pdf_to_markdown("input.pdf", "output.md")

https://example.com/v1 is a placeholder, not a working endpoint. Use a compatible service that actually provides the OCR model. See OCR configuration for other models.

Open output.md after conversion. Documents containing images also produce asset files; keep those files with the Markdown when moving or sharing it.

3. Create an EPUB

Reuse the configured craft instance above and replace the last line with:

craft.convert_pdf_to_epub("input.pdf", "output.epub")

Open output.epub in an EPUB reader. For title, author, and rendering options, see PDF conversion and translation. For installation or runtime problems, see troubleshooting.

Translation and reusable extraction

Translate books. Supply a chapter translator when converting PDF to Markdown or EPUB, or translate an existing EPUB directly. Translation uses a separate text LLM; OCR and translation have independent configurations. EPUB translation can replace the original text or append the translation for bilingual reading.

Create a translated PDF. Extract the content, translate it, and write the translation back onto the original pages. This workflow also needs Ghostscript and suitable local fonts. Check the resulting layout against the source and translated text.

Extract once, reuse later. Save a .pcex extraction file for subsequent rendering, translation, or processing on another machine. Reuse the configured craft instance:

craft.convert_pdf_to_markdown(
    "input.pdf",
    "output.md",
    extraction_path="book.pcex",
)

See PDF conversion and translation, EPUB translation, and the .pcex format reference.

OCR and runtime requirements

PDF Craft supports DeepSeek OCR, DeepSeek OCR 2, and Unlimited OCR, each with local and remote configurations.

The standard installation supports remote OCR. For local OCR, install the extra:

python -m pip install "pdf-craft[local]"

Local execution also requires a matching CUDA-enabled PyTorch build, sufficient VRAM, and model files. Models download from Hugging Face by default; you can also download them in advance and load them locally. Presets and requirements vary by model; see OCR configuration.

Language support depends on the processing stage. README languages describe documentation availability. Text recognition depends on the OCR model, and translation depends on the translator and text LLM. The EPUB lan parameter currently offers zh / en; see the API reference.

Documentation

Feedback and contributions

Report problems or suggest improvements through Issues. For conversion problems, include the package version, OCR configuration type, error logs, and a minimal file you can share publicly. Remove credentials and private content first.

Pull requests improving code, documentation, and translations are welcome. Keep translated READMEs aligned with the English version, including capability descriptions and examples.

If PDF Craft helps you, a Star helps others discover it.

Related projects

Wiki Graph can turn converted EPUB or Markdown books into structured summaries, chapter topology, and knowledge graphs.

License and acknowledgments

PDF Craft uses the MIT license. Third-party dependencies and selected OCR models retain their own licenses.

Thanks to DeepSeek OCR, DeepSeek OCR 2, Unlimited OCR, doc-page-extractor, pyahocorasick and the open-source projects that make PDF Craft possible.

Contributors

Thanks to everyone who has contributed to PDF Craft. Contributions to code, documentation, and translations are welcome.

Star History