Wanted: a lightweight open-source OCR engine, ideally as light as the OCR engine built into Windows 10/11 (apparently intended for mobile apps while MS was still trying to play that game), so not one of the current heavy deep-learning-based implementations, but also as good at handling color screenshots (particularly those with a mix of background/foreground color schemes) as Windows OCR. In my experience Tesseract fails on that latter point; would love to know if I'm just missing some trick.