doc2vision 0.1.1

Convert PDFs and images into clean, LLM-compatible image formats

doc2vision

doc2vision is a robust Python utility designed to convert documents and image files — including PDF, JPG, PNG, and TIF — into clean, high-quality, RGB images optimized for multimodal LLM input (e.g., image + text AI models). It handles low-quality scans, rotated pages, and multi-page PDFs with ease.

Perfect for use in OCR preprocessing, AI pipelines, or anywhere clean document-to-image conversion is needed.

---

🚀 Features

• ✅ Converts PDFs (including multipage) into individual high-quality images
• ✅ Supports JPG, PNG, TIF, TIFF
• ✅ Converts all output to standard RGB format
• ✅ Optionally auto-corrects skewed scans
• ✅ Gracefully handles edge cases like:
  • Low-resolution scans
  • Rotated or misaligned documents
  • Corrupt or unsupported files
  • Mixed DPI across pages

---

📦 Installation

pip install doc2vision

You may also need to install Poppler if you're using PDF input:

• macOS: brew install poppler
• Ubuntu/Debian: sudo apt-get install poppler-utils

---

🧠 Usage

from doc2vision import convert_to_llm_ready_images

# Basic usage
images = convert_to_llm_ready_images("example.pdf")

# With skew correction
images = convert_to_llm_ready_images("example.pdf", correct_skew=True)

# With resizing (preserving aspect ratio)
images = convert_to_llm_ready_images("example.pdf", resize_to=1500)

# Iterate over output Pillow images
for img in images:
    img.show()  # Or save, analyze, etc.

---

🛠️ Parameters

Parameter	Type	Default	Description
file_path	str	—	Path to your input file (PDF, JPG, PNG, TIF)
correct_skew	bool	False	If True, attempts to auto-detect and fix rotation
resize_to	int	None	If set, resizes image height to this value (keeping aspect ratio)

---

📁 Output

Returns a list of Pillow Image.Image objects, one per page/image:

[List[PIL.Image.Image]]

All output images are:

• RGB
• Preprocessed (rotation + optional resize)
• Clean and ready for AI or OCR pipelines

---

🤖 Perfect For

• Feeding documents into multimodal LLMs
• Preprocessing for OCR or Document AI
• Converting messy scans into standardized visuals
• AI agents needing consistent image inputs

---

📜 License

MIT License — feel free to use, extend, and contribute.

---

👨‍💻 Author

Built with 💙 by Russell Van Curen GitHub: @vancuren Website: vancuren.net

Changelog

0.1.0 - Initial release

0.1.1 - Updated project description and readme.