There are actually 7 or 8, even more if you count additional safetensors. But to answer you, the differences are how they were run and which model I chose (character vs object, tags vs caption, cross repetition # within vs epoch #, and I also ran a sd 1.5, sdxl 1.0, flex). There are also some minor tweaks between running's. Early versions had problematic images that ruined them under most prompts. Just trying to see what works. I will clean up my model page, when have consistency or at least descent outcome; leaving only what works. To do that I need to generate and experiment.