TextLens

OCR without the OCR complexity.
Read text from images and PDFs with one line of Python — on a laptop, a Raspberry Pi or a GPU server.

Website · Documentation · Examples · Changelog

Install

pip install textlens-ocr

Python 3.10+. No CUDA or PyTorch needed. The first run downloads a small OCR model (~31 MB) automatically.

Use it

1. Get the text

from textlens import OCR

result = OCR()("invoice.pdf")
print(result.text)

2. Export it

result.to_markdown()      # headings, lists, tables
result.to_json()          # pages, blocks, bounding boxes, confidence
result.save("output/")    # .json, .md, .txt and tables as .csv

3. Or use the command line

textlens ocr invoice.pdf                  # print the text
textlens ocr invoice.pdf -f markdown -o invoice.md
textlens inspect invoice.pdf              # which pages actually need OCR
textlens doctor                           # check your setup

That's it. For PDFs, TextLens reads the built-in text of each page directly and only runs OCR on scanned pages, so it's fast and exact.

Pick speed or accuracy (optional)

OCR(profile="edge")        # small devices (Raspberry Pi, Jetson)
OCR(profile="fast")        # default on most computers
OCR(profile="accurate")    # use the best model you have installed
OCR(model="glm-ocr")       # choose a specific model

Bigger models (for tables, formulas, complex layouts) need a GPU:

pip install "textlens-ocr[gpu]"
textlens models list
textlens models install glm-ocr

More features

Documentation

• Quickstart · Installation · Troubleshooting
• How it works · Models · API reference
• PDFs · RAG · Edge devices · Server · Docker
• Examples · Upgrading from 0.x · Changelog

Development

git clone https://github.com/Srevarshan05/textlens.git
cd textlens
pip install -e ".[dev]"
pytest

License

MIT. OCR models are downloaded separately under their own licenses (textlens models info <model>). See THIRD_PARTY_NOTICES.md.