this is a prelim sample of images testing the effectiveness of various things.
All input images are always the same, the different configurations are either
Different captioning levels of detail (Florence 2 large)
different prompt for llama3-70b to rewrite