>>57889989Yeah, echoing the other anon's points somewhat;
think about your prompt structure, I usually go style/quality -> character -> details, in your example tossing tons of really highly weighted hand tags at the very beginning of the prompt is very likely to cause issues with the rest of the generation
Most anime finetuned models seem to work best in portrait, I'd give that a shot. Using Euler shouldn't be an issue, but going past 30 steps with it is usually just a waste of time
And yeah Loras can have a big effect on your output too
I'd start with some simpler prompts with no loras to see if your settings and model are the issue, then start building the prompt you want and see what starts messing up the outputs
