ok I'm not a newbie, but this is worse than any other models before !!!
Can someone describe me the details and if all loras work, if it is more stable or not?
What i must choose for "Diffusion in Low Bits" for lora and does it make sense for an fp8-s model to use an fp16 lora ???
And which model type is faster: steps/it or it/sec (not if its only need 4-steps or 30)
DEV, 1.S, FP8, BNB, NF4, Hyper, Schnell, GGUF, 4-STEP, 8-STEP, UNET, CLIP, VAE,,T5XXl ...
for which model i need which clip,vae,t5xxl ??? -> it give some times 4 versions ...
and importand ... how much VRAM 12, 16 or 24 i need ???
PLS CIVITAI provides clear categories !!!
...all these stuff
WHO write an articel ? ^^
for me the best so fare in all cases (speed,lora)
hyer (12-14 Step):
https://civitai.com/models/638187?modelVersionId=819165
or normal (25-35 Step)
https://civitai.com/models/638187?modelVersionId=721627