use miaomiaoharem checkpoint if your result not as good as the example i gave,

testing training in civitai, but damnn became one of my favorite model so far,
choose the epoch yourself,
epoch 4 stronger but sometimes kinda overfit? like the result doesnt follow the prompt, but sometime still good,
epoch 2 more soft , the prompt still following, both are good xixi,
but me personally epoch 2 is enough

euler a with normal, or er sde with sgm uniform, 30-50 step, if using turbo 15 step fine most of the time,

pls support on my ko-fi if you like model https://ko-fi.com/rastaai . i love making those lora but since i makin it on cloud gpu it'll be really useful for training, thx.