Docling is great for PDFs https://github.com/DS4SD/docling but if the input is really only images (in PDF) than cloud AI based solutions (like latest models from Google) may be better.