Quoted By:
Saw the new embed making guide and came back for a rematch with that anya embed I made a while back, even with mostly the same dataset (it has 3-4 more pics) it seems to have come out better? It doesn't mess up the eyes anymore, and the hair looks more anyaish at least. It's a bit better with different outfits but still sometimes hard to get her to wear things that don't look like her default black dress, which I suspect is the result of ~90% of the training data being that outfit. Not like I really have a choice...
One thing I thought was interesting is that the new one always puts her on a plain background while the old one kinda changes the background depending on the outfit, even with no background prompting.
I still have no idea what I'm doing but the guide is good I think
