Left - normal lora and onetrainer, right - lora-plus and sd-scripts.
There's not much difference. looks like lora+ requires lower lr and i set it too high
https://litter.catbox.moe/5z9xgt.jpgfuta comparison:
https://litter.catbox.moe/pqweme.jpgrandom non futa:
https://litter.catbox.moe/ybqrm3.jpghttps://litter.catbox.moe/35h96a.jpghttps://litter.catbox.moe/cm5tqw.jpg>>75726288If hair color and 'multicolored hair' are tagged, having to use full tags is normal
>How do you decide if the bake is under, lagom, or overcooked?I just to try guess how many epochs is needed and look at images
If this was not a diffusion model, you could look at validation loss and stop when it starts going up, but looking at loss when training SD is pretty useless
