This is handy — I've hit this exact issue prepping documents for LLM context. How's the accuracy on lower quality scans?

[dead]