how is the GenAI accuracy compared to AWS Textract? Also I am not sure the libraries used by OpenAI get details from image input. I think that understanding page layout of image is very important, which might not be the focus of OpenAI.