>>58604386as for the settings that exist in the first set of Dynamic thresholding metadata but not in the stanky old maririn from March
TL;DR they added a fuckton of options to it - I have not touched any of them, they are all at default. when I touch them, shit breaks, kek
also I probably misspoke, been too long. the really high clamp percentile is what is doing most of the wash out look, you'll get it regardless of scheduler>>58605761if you're just looking for the benefits people expect out of it (fingers, definition, etc), you use cfg 8 most of the time? set cfg to 16, tell it to mimic 8. and then stick with:
>Threshold percentile: 95, Mimic mode: Power Up, Mimic scale minimum: 4, CFG mode: Half Cosine Up, CFG scale minimum: 4, Scheduler value: 4to start
the different schedulers change the rate at which cfg gets changed per step. half cosine up = follows a cosine curve up to HALF the set value. power up = follows a power curve up, etc.
I think half cosine up on CFG was the important adjustment to get it playing nice with DPM++ Karras samplers. if you use e.g. regular cosine up, it will fry. I must've messed with different combos for hours when it first came out, brain rusty tho. especially since Restart + Rescale work a bit differently...
if you want it like that Maririn, turn the clamp up closer to 100. play with it, the further up you get past 95, the more it starts to change with very minor adjustments. like the diff between 93 and 95 is smaller than 99 and 99.5.
>>58605781my body is fkn ready
