It's Filene, but now in LoRA form. A bit of an expanded dataset and better tagging than the TI embed from earlier. You can use both together if you like. I took particular care in trying to train on as many from behind and from side images as possible so that the tail looks good from any angle. You still get the occasional detached or misplaced tail when prompting not from front, but you should be able to roll into a good looking tail pretty easily. As always, trained base NAI with FULL TAGS. That means that the costume was always prompted during training, so you MUST TAG IT AS IN THE EXAMPLE IMAGES.A few training images were with her bodysuit costume, but not enough to get it really accurate. You can try it if you like though - see example.

Files with SHA256: 9cb1e681c85786f969ce6101cb78f5a806d46e35db6f4cf79ffb00a189244f66