Lora for Nanami from doa. This was trained with 饭特稀_v0.8 which you can find on civit. The lora should work ok on similar models but it looks best on 饭特稀_v0.8. Trigger word is "nanami".

Her highlights seem pretty subtle from the lora but overall I think it got her face pretty well. Some of the training images had red highlights so you may get see that rarely too.

Also highly recommend EasyNegative which my sample images are using.

If you get bad looking faces at lower resolutions try inpainting only the face at higher resolution or using hires fix.