What's your workflow for using Zimage base?
is it different from what you use for Zimage turbo or is it the same?
Hey!
You can check the browser's samples for ZBASE, they are generated on BASE mostly so you will see the workflow if you download the image and drop it to comfy :)
Hi, I don't know why I can't get your ZIB-trained characters to work with the ZIT workflow, but it's strange because they should theoretically be compatible, and I often use ZIB-trained LoRa on ZIT. Maybe just typing sks woman as a trigger on ZIT isn't enough? It doesn't work even if you use the full LoRa syntax or significantly increase the LoRa strength. In that case, the facial features improve, but if I increase it too much, the image gets ruined.
You need to up the strength of those zbase loras on turbo (to even 2 or 2.2)
this is due to issue with the optimizer used during training, now people know more about this issue. I will definitely want to train on base in a way that it works fine on turbo too, this is why i paused training new zbase, I need to figure out good params.