I loved the iDOLM@STER / Idolmaster Million Live&NIPPAN collab PR cards… so I made LoRAs on them!
Due to the dataset really being 10 copies of the same image, it will generate basically the same thing... same pose, same face, same facial expression, same manga books being carried... even the hand pinching her glasses.
Most of the times the hand pinching the glasses looks correct (although sometimes it comes out wrong because it's AI), however the hand carrying the book not really.
If you want more freedom when generating, it's better to use it at a lower weight.
