Will adequate prompting or the use of solely realistic models produce realistic results for your LoRAs, or is this 2.5D/cartoonish look baked into all your models by default (as your samples indicate)?