This is a more realistic version of Semi-realistic style. Once again, avoid any models such as DucHaiten or JunkJuice which have very distinct styles and will less likely have an effect on the LoRA.
To get desired results are much more difficult in contrast to my other LoRAs.
To optimise best results: Use models from the preset, such as Hakura to ensure that you get most out of this LoRA.
Better results range 25-30 sampling steps, I typically use 25 since I'm cheap and constantly run out of points. However, Hisoya claimed they couldn't identify much a difference between 25-30. Hisoya did and said 20-24 provides alright results.
For CFG scale, this will constantly revolve around what model you use. Me and Hisoya used Haruka model as that is what the model the LoRA was trained in. For Hakura, I only use 5.5 but Hisoya says you can range between 5-6, not higher. She says 6.5 provides weird results.
Hisoya also stated that unique hair styles such as man buns tend to lack realistic hair, but short and long hair are fine.
I only tested out one sampling method which is Huen in Hakura for the LoRA and it gives an oddly scratchy look. Not that it's unappealing as I believe the shading is better to me in my opinion, however, it doesn't have the style I desire. Hisoya did a bunch of DPM++ sampling methods. However, she was dissatisfied with the results, I personally thought they were okay.
That's all the advice I can provide in the top of my head. The best results, of course, are portraits and close-ups, however, it can do appealing cowboy-shots.
