Added a v2 of the finetuning hypernetwork which was more aggresively trained and seems to yield higher quality images

added finetuning hypernetwork trained on main dataset without face covering helmets