Changes in V2 Loras
What did you change for the V2 Loras? I feel like I am getting worse results for the V2 loras than the V1, but maybe I am not using the right settings. Do you have suggested sampler/scheduler?
I don't think it's worse, I believe it looks cleaner. If you combine the 1v and 2v, you might see some nice results.
I'll try combining them. Using the V1 billie I am getting much sharper results and the V2 seems to have a much more blurry face.
V2 for Billie was a special request from a special dataset (wwafawdwg) from this https://en.wikipedia.org/wiki/When_We_All_Fall_Asleep,_Where_Do_We_Go%3F
But yes, those are also good to combine together, however for Z Image Turbo you need to remember to not go over 1.2 or you will see a big degradation in quality and artifacts start to pop
Btw, there is a v3 version trained on 200+ images, that one should be superior to the other two :)
Using the V1 billie I am getting much sharper results and the V2 seems to have a much more blurry face.
Yes, that was the aesthetic for the image set
Hi Malcolmrey,
thanks for all your LoRA. I have a quick question: I tried using zimage_jessicaalba_v1 with Z-Image Base, but it seems to only work with the Turbo version. Was it trained specifically for Turbo?
Hey @Jubien3132 - yes, sadly all zimage (trained on turbo) won't work on base. But I also trained a bit (not many) on z base (they have zbase_ prefix)
But zbase can be used with turbo