>>38357730>spoilerpicrel
Im on the fence. On one hand training every outfit might be best for a complete model but those are huge and not as shareable.
Training an embed for every outfit feels like a fucking hassle, either way youd need more data and vectors to get the outfit correct. But that all goes down the drain once you try to prompt away the clothes.
Im not the author of the Sana embed so I cant speak for it. I can only voice my own critique from my 1 month experience I have training various embeds myself. My early ones are kinda garbo too. The turnaround was on the Fauna embed really.