>>87682824>I'd still think this would probably influence the styleAny LoRA will influence the style, but the effect is so small it doesn't really matter. When it works correctly.
It's just to get the base output to look softer and better. I'm not a fan of flat, overly 'anime screenshot' type gens.
>So you're training them yourself?Nah, I tried a few times but it never really worked out. I just grab, test and combine LoRas.
>Can I ask why you're not doing a concept lora for what you're trying to improve then? So, images of different artists doing the thing you want better, making it not aligned to a specific style.I've seen a few LoRas that supposedly improve backgrounds as a goal, and I didn't like a single one of them. Either they're trained to only work for some specific backgrounds (eg. mountains) or they're trained to only work for some specific backgrounds, but don't tell you.
I don't think a concept as broad as 'Do what you were doing before, but better' is doable intentionally.
>cause if that's what you've always done, it obviously worked for other models.Yup, been doing it since 1.5. Not the same LoRas obviously.
