Style-LoRA of the artist Aurahack.
LoRA created via Pixai LoRA onsite-creation. Used 82 Images on high/mid resolution, partly from rewards.
Weight at 1 seems to be working good, but honestly the style works hard even at low weights. Like Aurahacks artwork the style tends to overwhelm anything in its way, but it's fun to use :). All preview images are done with Cetus-Mix (v4.0). Highres-Fix recommended!
You don't need to use all trigger words, but it might help. Also play around with weights!
Description
Initial version trained on Pixai
FAQ
Comments (2)
What makes you decide what model to use with which LORA? I notice you usually use galena.
Thanks for asking! Since I train my LoRAs on Pixai I have a limited selection of base models over there and tend to use their anythingv5. After that I run x/y/z plots to get an idea which checkpoints lends itself best to the original imagestyle. Since the styles I published so far tend to be less anime-esk I look for checkpoints that aren't...well...anime-first ones AND also work well with what I aim for. Galena Redux simply turned out to be a very good choice most of the time.