Tiona from danmachi

I genned images locally, face detailed most.
40 steps,
sampler: euler
scheduler: simple
cfg: 4

seems to work, though I may train again with a larger dataset.