Does fine-tuning normally take 15000 steps to fine-tune a more normal human? I am guessing it might be more since it's so distinctive.
Also, if you wanted to be able to generate normal images of real people also, would it be sufficient to identify them as Minecraft characters in the prompt? Or would you need to also include some regularization images also in the training data?
Also, if you wanted to be able to generate normal images of real people also, would it be sufficient to identify them as Minecraft characters in the prompt? Or would you need to also include some regularization images also in the training data?