Using images generated with V1, V2 now has expanded training data and destroys backgrounds less as well as does not put the head out of frame for burlap outfits.
V2 has a more difficult time doing a single burlap item unlike version 1 that can have a normal and a burlap item at the same time.
If asked to improve it more would try to make it be able to do more half burlap half normal items easier and include more hair colors in data set, at the moment tends to lean towards black and brown. This model was trained on WarriorMama777's abyss orange mix 3 from huggingface.