Type
LORA
Publish Time
2024-07-21
Base Model
SDXL 1.0
Trigger Words
DFxnk design
Copy
Version Introduction
Redid captioning and training dataset. Also trained at different rank and alpha, steps.
This version works best with dpmpp_sde_gpu karras, but also does decent with euler normal.
I use it mostly with my unstableJugg checkpoint merge.
It a bit more responsive to prompts, has more control, and less randomness, or rather not listening to you at all. It can still create craziness, and that is somewhat the point of the style.
I'm not sure that I will be visiting this one again since it does what I want for the most part. I also trained this back in February, so I don't remember what the steps were.
