stable-diffusion
stable-diffusion copied to clipboard
Question about additional steps
I've noticed that adding additional steps often changes the model output completely. If I run the same seed/settings through with 10, 20, 30, 40, 50 steps I get different non-noisy results instead of progressively less noisy versions of the same result. What's going on under the hood here and is there a way to directly manipulate it? For example, could I have it do 50 steps with the 30-step process to get a super-denoised version of the 30-step result instead of a completely separate result?