Quoted By:
I tried making an anya embed and it's so bad I don't even want to link it here. I suspect it's a problem with my training data selection rather than a setting I forgot somewhere because it generates pictures that still look ok as images but only very vaguely look like her.
Any advice for making embeds? Can deepdanbooru be trusted for tagging or is it a better idea to tag the images yourself? How many pictures is a good number to train on?
