option to output all intermediate steps up to the sampling steps amount entered as images

[OWKenobi]

Hello,

for some images, the result changes drastically from one sampling step to the other, and one time, 14 is perfect but 15 is a desaster, sometimes the sweet spot is on 20, and so on. Now, instead of having to manually produce all sampling outputs for comparison, which takes a long time, and the computer re-calculates all the early steps all the time, could you add an option so that all intermediate steps up to the sampling steps amount entered, will be outputted as images?

This way, if you select 20 sampling steps, in the time of only 1 generation you would create 20 variations of your image.

[Baku-Rue]

Was just getting on here to request this same exact thing. Being able to get even a range of pictures from a set of steps would be amazing. I was considering trying to loop a prompt or even use the multi prompt feature, however you can only set the steps by the slider in the GUI and not in the actual prompt itself.

[Baku-Rue]

I actually just found this repo from another post asking for something similar. Its a little janky but it does exactly what I was looking for.
https://github.com/404james/stable-diffusion-grids

[OWKenobi]

Hello Baku-Rue, awesome, this looks like an automated batch maker. Thanks for the tip, I'll have a look! Sadly, it doesn't catch the idea that, in theory, the diffusion would have to run only ONCE to generate ALL steps, which is extremely faster. Am Fr., 2. Sept. 2022 um 22:11 Uhr schrieb Baku-Rue < ***@***.***>:

…

I actually just found this repo from another post asking for something similar. Its a little janky but it does exactly what I was looking for. https://github.com/404james/stable-diffusion-grids — Reply to this email directly, view it on GitHub <#238 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABG62LXKKCAKAN6K5IXBWQDV4JNOLANCNFSM6AAAAAAQDDRWQY> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[Lumoria]

This is much needed! Ideally, it should be implemented like how it worked in disco diffusion. Have the diffusion process output an intermediate frame for every X number of steps, X being a variable set by the user.

[cobryan05]

This would actually be a fairly simple addition to Sygil-Dev/sygil-webui#611 . I might look at it tonight.

[cobryan05]

Added to PR