Visual Understanding with GPT

Which image input method is best for a batch processing pipeline that analyzes the same set of product photos repeatedly?