Blame for README.md

---
title: litert-ocr
emoji: 📄
sdk: static
app_file: index.html
pinned: false
license: apache-2.0
short_description: Photo, scan or PDF to text, Markdown and JSON, in-browser
models:
  - litert-community/PP-OCRv6-Small-LiteRT
custom_headers:
  cross-origin-embedder-policy: require-corp
  cross-origin-opener-policy: same-origin
  cross-origin-resource-policy: cross-origin
---

# litert-ocr: photo, scan or PDF → text, Markdown and JSON

Choose a photo, a scan or a PDF (or drop one, or paste a screenshot), and [litert-ocr](https://github.com/john-rocky/litert-ocr) reads it in this browser tab: every text line with its box, a PDF's page with a frame around each block, Markdown with headings and tables, and JSON. It runs PP-OCRv6 Small ([litert-community/PP-OCRv6-Small-LiteRT](https://huggingface.co/litert-community/PP-OCRv6-Small-LiteRT)) on LiteRT.js. Your files are not uploaded. The page downloads the library, its LiteRT.js runtime and pdf.js from jsDelivr and the model files from Hugging Face, once, and the browser keeps them.

The same page without the Space's frame: https://mlboydaisuke-litert-ocr.static.hf.space/. Inside the frame, Chrome on a Galaxy S26 read more slowly.

- This page's code: [demo/](https://github.com/john-rocky/litert-ocr/tree/main/demo) in john-rocky/litert-ocr (the same page is on [GitHub Pages](https://john-rocky.github.io/litert-ocr/))
- The library: `npm install litert-ocr`, [README](https://github.com/john-rocky/litert-ocr)
- The examples: [examples/SOURCES.md](examples/SOURCES.md)