Quoted By:
I am not giving up on Sana.
I've been training an embed with my CPU because I am apparently insane enough to try. It somehow works, even if it's slow as shit.
While it's still far from optimal, it's starting to add some more Sana-like details. (VAEs are off because training, that's why it's extra wonky)
I'm using the same seed because it's different enough to serve as control. I'll see when it's about 5000 iterations, if it can make something resembling Sana more from this base, I know it's more or less working.
So far it goes schizo with the clothing (turned the dress into a veil here) but it's starting to notice "add more balls to the hair" and rounding the eyes more.
I wish Sana drew herself more, her self-portrait is one of my favorite pics of her, but I don't know how to make it weight more in the embedding outside of repeating it a few times, but I feel that'd be bad.7GP84
