/vt/ - Virtual Youtubers

>>60710764
It's slightly more complicated: you plug it into your a1111, convert your model (takes few minutes, but you only need to do it once), convert each lora you're planning to use(also few minutes) and THEN you "run things normally"

>>60711589
on windows i got the following:
512x768: batch 1: 22 it/s -> 30it/s, batch 4: 6 it/s -> 9it/s
hires fix 1024x1536:
batch 2: 13 / 1.8 it/s -> 17 / 2.5 it/s

50% gain, basically.
On linux it should be even faster, because of triton