//=time() ?>
Bonus mini test - with DDIM, I get to play with the eta setting! The eta values are in the image descriptions.
PLMS
More varied results on certain settings, but also garbled like the LMS set at low steps/high CFG. Fast, but not the best.
LMS
Good for speed, not for high CFG or low steps. Too much distortion, but fine for general use. (repeat images from first run, with higher steps)
EULER
My second best (after DDIM) for speed and low step results. (repeat images from first run, with higher steps)
HEUN
Also twice as long to process. Also good for low step generation, but maybe not worth it for more?
DPM2
Some more interesting low step outcomes, but twice as long to process.
DDIM
Generally faster (see earlier thread) and cleaner overall, with nice details at low steps.
#StableDiffusion Following in @proximasan's steps (pun?), I've put together a comparison of steps vs CFG scale. So far, using two sampling methods, may try some more, but the results were not as interesting as I expected. This first set was the k_lms method.
@EErratica I did something similar. Love how well this works! Although, apparently my bad grass turned into a golem, lol.