Ksampler Advanced - Denoising Calc版本v1.0 (ID: 381564)

Lots of beginners ask "why there's no denoise function" in Advanced Ksampler. To answer the question - is because denoise in itself is a calculation of steps. ie... a 0.75 denoise basically means in a 20 step flow... everything starts from the 5th step.

50% denoise? starts from the 10th step.

it's a calculation you can implement yourself : denoise = (steps - start_at_step) /steps

i've created a "relatively" simple workflow to show just that.

this essentially allows you to handle advanced ksampler the way you would handle a normal one.

.... "but why wouldn't i just use a regular ksampler then!?!?"

good question. basically control. when doing timestepping, you'll want to understand and control the denoise via step count... not % of denoise.

this won't solve that conundrum for you... but it will get you closer to understanding the process.... not to mention get you used to using the advanced Ksampler in your workflows.

this is primarily useful when doing a second pass or running a refiner

......"but the ksampler and advanced ksampler are slightly different!!!"

yep. rounding errors. steps are integers... but % is a float. this is close enough for showing how to use it. you can mess with the formula yourself and try to make it more accurate. i suspect for 99.9% of the use cases this will work just fine.

installation should be as simple as throwing the json to your workflow... manager > add missing nodes.

i think there's 3 custom nodes used, the pythongsss /comfyui logic/and Mikey nodes... there's other ways to do this, as the end of it it's just math expressions.... but this was easier for me to implement.

描述:

训练词语: