# Pikabaka
LoRA trained on images from ai artist: pikabaka. (RIP ?)
The Kroma variant is the best one for now imo.
## Kroma
The Kroma version is better in every way compared to the Krea2 one, I invite you to use Kroma instead of Krea2 if you plan on using this.
It might even be better when applied to Krea2 than the Krea2 version if you care about prompt following (style will be worse)
## MOSS-sft
Not quite satisfied with how this one turned out. I have a few ideas on how to improve it though. (Redoing the captions would be a nice start)
Civitai inference also seems worse than ComfyUI with same prompt and same settings ? Idk you will have to experiment.
Description
Training Settings: batch size 4, 1024px, ~175 images per epoch, optimizer is SinkSGD_adv



