DevilHS Style Lora.

Great style. Lora tends to output very top heavy results, but then again that is part of the artist's particular style.

I made two versions with the same dataset:

1) Illustrious-based (IL)

2) PonyXL based (PonyXL)

The IL model, its merges (and NoobAI) all have the style baked in and it works fantastic in my opinion, but I have been using different checkpoint merges (like the NTR mix or the WAI mix) and custom LoRAs occasionally, so the raw, vanilla Illustrious and NoobAI experience may not be as good.

In my humble opinion, IL won this one. And that seems to be a trend when it comes to following prompts especially. Must be some of those ingeniously hidden tags, strings, hashes of Pony that make the magic. Alas, I do not know them.

*missing text*