ComfyUI: no GPU? a workflow for SDXL | "take a while and listen" <-decard cain, Diablo

hi folks,

this is just a short one to introduce my modified worklow.

i am using different sdxl models, starting of with the sd_xl_base_1.0, but its more fun to try out some models. also using the sdxl_vae.

it is very slow without a GPU, but the results are convincing.

the point is, with 3 sd_xl_refiner_1.0 and (adv)Ksampler, running 3 times smoothes and refines incredibly.

in case you have to much time, or let it run while you are not at your

device (mine: amd ryzen5 3,6ghz + 48gb ram),

with a resolution of 1024*1024 it takes (with 16 steps at base model and 3 times 4 steps with the refiner) about half an hour.

to time i prefer a model which works with lesser steps as usual (forrealxlV10_v10) to speed things a bit up. no matter what i do, with this workflow 1 iteration takes about 60-70 seconds...

I'll try it with FLUX in the future, but i'm afraid of how long it then takes... o.O

i would be happy to receive feedback and perhaps a few tips on how to speed up the process.

have fun!

(and please forgive me the mess)