pix2tex (LaTeX-OCR)

Use it when

• You have a screenshot or image of a printed math formula from a paper and need the corresponding LaTeX source instead of retyping it symbol by symbol.
• Generic OCR engines mangle equations into garbage characters and you need a model trained specifically to emit LaTeX markup from formula images.

What it solves

Not the fit when

• Handwritten formulas (only partial, experimental support per the README TODO)
• Very large or zoomed-in high-resolution images, which degrade accuracy despite the resolution-prediction preprocessing step
• General text OCR of full documents; it targets isolated formula crops
• Unverified use: the README warns to always double check the predicted LaTeX
• full-page document ocr
• pdf table extraction
• scanned document layout parsing
• speech or audio transcription

Install

pip install "pix2tex[gui]" (requires Python 3.7+ and PyTorch; model checkpoints download automatically)

Invoke

Run `latexocr` for the GUI screenshot workflow (predicted LaTeX is rendered and copied to clipboard), or `pix2tex` on the CLI for images from disk or clipboard; in Python: from pix2tex.cli import LatexOCR; LatexOCR()(Image.open('img.png'))

Alternatives

No reviewed alternatives recorded yet.