Extract text from scans and photos
Optical character recognition for scanned pages, photos of documents and screenshots. Works on images and on PDFs that are really just pictures.
Questions
How accurate is it?
Very good on clean, straight, well-lit scans of printed text. Accuracy drops with skewed pages, low resolution, unusual fonts and handwriting.
Why is the first run slow?
The recognition model has to download once, which is a few megabytes. After that it is cached and subsequent pages are much faster.