Use weight of around 0.67 for best results.

LoRA trained on SD 1.5.

Images trained mostly on ingame screenshots. It was very difficult to find good images of Leliana without her hood. I used screenshots from across all games and some promo art. This LoRA is more flexible than I expected despite the limited training data. Her hair color and clothing can be changed pretty well. Unfortunately, the lighting is usually a bit dark and orange/red (which is how the games were).

Images uploaded were generated using GalaxyTimeMachine Photo model. However, also works well with DreamShaper and chilloutmix.