"Your generation is as weak as your aesthetics.
Failure is your destiny.
You disrespect yourself, and your creator.
You are made of stupid (settings)."
-me when
Well, just some collections of my experiments.
75vp3-000013 :
It was supposed to be the upgraded version of my already-released CanKing LORA ( which you can find in my profile ). And enhancing several issues with 75 Image variant 1 (thick lineart) and 2 (saturation). But ehhh..... this abomination cook happened with me experimenting DAdaption.
This model stopped training at 13th epoch as I stopped it.
No activation word.
lemon20 :
This is a LORA for someone called Lemon Nishikawa from Hope You're Happy, Lemon. When it is still in its early training, everything look fine so far. - And then this came out when it reached 20th epoch. The LORA is very rushed that I didn't properly clean up the dataset, and just slap it into autotagger. Plus that was my first time creating character LORA. So yeah, it actually didn't replicate the character, more or less.
"lemon, silver hair, white hair, long hair," is the activation word for it.
My definition of failed LORA is NOT something that could be generate something acceptable to look, or it is just some minor artifacts here and there. Here is something you can ruin the entire image for you. ( I may contradicted the thing other way around if you actually turn it into lower weight but still. )