Repeating generations?

It seems like civit has just started repeating the same images when using the same prompts, with minor changes if you tweak one or two prompts, but what if the results you end up with aren't even close to what you're aiming for?

I'm confused by this move, as it basically removes variance in results, and strips back potential a hell of a lot.

Could you give some more details on your input?

If you use the same seed with only slightly different prompts then the changes will be minor as well. If your clip skip is really high or the cfg scale is really low that could have similar results.

The variance comes from the seed and the other two affect how closely it conforms to the prompts. So make sure that the seed is set to random or you manually change the custom one if you want. CFG at 7 and Clip Skip at 1/2 (realistic/anime) is the baseline.