Great LORA and dataset!
#3
by fawogin598 - opened
Thank you very much for sharing your dataset.
I'm learning how it works and would love more info if you're willing to share.
Did you only use images with "style" in the prompts to train this LORA?
I don't get how the lego example in your sample image has (almost) correct hands, since the 4 "lego" images in the dataset have wrong or no hands.
Didn't you inverted images in the first example about crying?
Sorry for the delay, it generalises style transfer based on the dataset, not learning from the specific examples but rather the actual task (though examples w/o flaws will help obviously)
peteromallet changed discussion status to closed