Maybe make it possible to use https://github.com/JaidedAI/EasyOCR? Instead of tesseract it is a "AI" /ML based OCR AFAIK. It is e.g. used by https://github.com/ocrmypdf/OCRmyPDF-EasyOCR