>>105919958
>>105919971
Right, I did train on illustrious for a while, (though, to be honest, I haven't checked if those loras worked just fine on base illustrious.) but I was testing the loras on Illustrious merges and they always looked "weak" or not right.
Then I moved to NoobAI 1.1 because I heard that it was a better and more knowledgeable Illustrious... I was still testing the lora on merges (NoobAI 1.1 merges this time) and they naturally looked "weak". I had to increase the strength of the lora to "1.8" once to get something that looked somewhat like the style.
But, when I generate on the base NoobAI 1.1 ckpt, the lora actually works. I don't have to decrease or increase the strength for it to look how it's supposed to.
I guess I'm asking if that's normal for you all who are more experienced in baking loras for sdxl/illus/noobai models? I've made a few loras that go well into the 40s epochs and they look completely wrong on the "better" models but just fine on the base checkpoints.
