Skip to content
This repository has been archived by the owner on Nov 9, 2023. It is now read-only.

Steps vs. timesteps #9

Open
djdookie opened this issue Aug 1, 2023 · 1 comment
Open

Steps vs. timesteps #9

djdookie opened this issue Aug 1, 2023 · 1 comment

Comments

@djdookie
Copy link

djdookie commented Aug 1, 2023

Hi. Your extension is really helpful in Auto1111.
But like described in https://github.com/vladmandic/automatic/wiki/SD-XL#using-sd-xl-refiner we should be able to define when the refiner takes over from base depending on the remaining noise (relative). Like 0.8 (80%). Not based on the absolute scheduling steps.
This doesn't necesserily mean that the refiner takes over at 80% of scheduling steps!

Quote:

Note on steps vs timesteps. In all workflows (even with original backend and SD 1.5 models), steps do not refer directly do operations internally executed. Steps are used to calcuate actual values at which operations will be executed. For example, steps=6 roughtly means execute denoising at 0% -> 20% -> 40% -> 60% -> 80% -> 100%. For that reason, specifying steps above 99 is meaningless.

Since the schedulers behave differently with denoising, this could be very helpful to get more consistent result with different schedulers.
So it would be nice to be able to have a slider for Refiner start: 0.0->1.0, default 0.8 (as proposed by Stability AI).

@wcde
Copy link
Owner

wcde commented Aug 1, 2023

I did some experiments, timesteps works with normal samplers, but not with Restart. It has a second denoise peak and it is very difficult to find good value so that sampling does not break.

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants