2024-02-12 UTC
# 
capjamesg [tantek] The book system works as follows: detect and segment all books, to the pixel level, with a machine learning model. For each book: crop, rotate to the left by 90 degrees, send to GPT with Vision for OCR (GPT has great OCR that's easy to set up and is fast; could use the Apple OCR API on Mac but that would limit the script to macOS-only), look up book with Google Books API, create SVG and overlay it on image.