Show HN: Textsnap – CPU OCR in <1 seconds/page, with VLM fallback for hard pages

Category: devtools

Tags: ocr, cpu-inference, onnx, vision-language-model, cli-tool

Score: 6.3/10 (Innovation: 5, Technical: 6, Documentation: 8, Utility: 6)

Textsnap is a CPU-based OCR tool that combines a fast lightweight model with a vision-language model fallback for difficult pages, offering offline, single-command text extraction from images, screenshots, and webpages. Its dual-engine approach and portable model setup make it practical for local, privacy-preserving OCR.

Target audience: backend developers, devops, data engineers, end-users needing offline OCR

Repository: https://github.com/kouhxp/textsnap · Python · MIT · 195 stars

View on Hacker News