>>77120645>Hyper is a weird one where it doesn't work with animagine for some reason.It does work, that was from animagine. I just had to set the lora model strength to 0.2
(and CFG 4, euler ancestral or dpmpp sde, simple scheduler for best results imo)
The CFG is supposed to be in the 5-8 range according to the model card but going higher than 4 seems to introduce a ton of noise.
Here's the same prompt with trifusion, 8 steps for the initial gen and 4 steps after upscale. Takes 45 seconds on my 2060 (half of the time being spent on loading models)
