stable-diffusion-webui
stable-diffusion-webui copied to clipboard
AUTOMATIC1111
Separate sampling steps for Highres. Fix
Is your feature request related to a problem? Please describe. When Highres. Fix is used the sampeling steps is identical on the first and second image produced. I am wondering if it would be more useful if the steps would be separated.
Describe the solution you'd like The ability to set the sampling steps individually. For example, 50 steps in the first lowres version and 130 in the second highres version.
Describe alternatives you've considered Creating a lowres txt2img image then send it to img2img and use different sampling steps
Additional context
Would this be related to https://github.com/AUTOMATIC1111/stable-diffusion-webui/issues/983 ?
I implemented that but I don't have a clean patch atm. It works well but you don't save that much performance. For example on my 3090 I have 15-18 iterations/second in stage 1 and 1.8 iterations per second in stage 2 at 1MP res. So 10 steps you cut in Stage 1 is as much as 1 step in stage 2, that's why the performance change only becomes relevant when generating hundreds of images.
I would also like to be able to determine the resolution of the first step, but I wont open a new feature request just for that
Would this be related to #983 ?
Yes, I think this feature will make Highres. Fix more useful. Also, if we could change the sampling method individually that would open up new possibilities with same seeds.
I was about to request this myself. When using a large step size for the final render, I don't need a lot of steps to get the composition of the lowres-pass. Either a cap, percentage of the second pass or an absolute number of steps would do the job.
Unless I understood incorrectly, this is now implemented in the latest versions. Consider closing?
