Ohhhh yeaaahhhh!

For anyone wondering, it is recommended to set the hyper LoRA to a weight of 0.125

How big are performance improvements?

I've been using this to sort of simplify compositions and use less steps to get usable results for anime work and its been great however, I am a bit bummed out that there seems to be some sort of banding of solid colors, mostly noticeable on dark blues.

IDK how it supposed to work. Without any lora I can generate ok images in 16 steps.
With this lora generation time is %30 more. It actually makes inference slower.
If I set 8 steps, images look bad.

I'm an amateur who has just come out of the novice village, and many models want to try boldly, but how to use them reasonably is indeed a headache~! The technology and model are updated too quickly, I learned to adjust 456, and I also tasted the benefits of retraction, but this process is too difficult for me, thanks to the contributions and help of the vast majority of the authors of station C.