Dont get me wrong - i love this implementation of SD - BUT ....

[OleJ1964]

I compared the images with other online resources (and image databases) - same Prompt seed, sampler and so on. The images from my own GPU and Automatic111 code produces very nice images - but imo. nowhere as detailed and good as i see else where (Like Here https://playgroundai.com/create) but many others, what (if anything am i doing wrong or not understanding here ??

I am not complaining - but have this strange feeling that I'm doing it wrong - the reason for sharing prompt and seeds i imagine is for others to be able to make comparable images.

[youmebangbang]

Use more steps (200+) and euler A sampler

[OleJ1964]

Use more steps (200+) and euler A sampler

I actually tried to push the steps, dont make much difference, and the online examples i tried uses the steps i tell it to. So it should be the same - but its not.

[ForsakenRei]

Since we don't have the backend code for most of the online implementations, we don't know if we set steps to be 20 the model will actually receive 20, depends on the implementation it can be 200 or 2. Another example is token limit, sd has a limit of 75(or 77 I forgot) but NAI implements theirs to be 225(with some tweaks of course).

For certain promot set I get worse result on local but I usually get equally good or even better results on most of my trails. More steps are better on the lower side, but once you pass 100 I guess there won't be too much difference? It really depends.