Closed johndpope closed 3 months ago
the way the training "works" right now is nuts. this is better direction from X-portrait - that is synthesize cross-identity image some matches pairs to reduce the training time. this is is SD land.
https://ar5iv.labs.arxiv.org/html/2403.15931
the way the training "works" right now is nuts. this is better direction from X-portrait - that is synthesize cross-identity image some matches pairs to reduce the training time. this is is SD land.
https://ar5iv.labs.arxiv.org/html/2403.15931