This is an attempt at training a LoRA on Zelda's traveling outfit in Breath of the Wild (and Tears of the Kingdom). The costume comes across pretty well, however (perhaps due to the relatively low diversity of characters wearing it in the training data) it kinda tends to zeldafy the characters that are supposed to be wearing the costume.

The flipside of this is, that it's kinda pretty good at making Zelda, so give that a try if you want.

You can see the issue pretty well here:

I'm gonna release this for now since it kinda works okay-ish (seemingly a lot better with models that have a very well-defined idea of what the characters heads are supposed to look like), but your mileage may vary.

My next step will be to split up the dataset and train the model on multiple costumes to hopefully solidify the understanding of where Zelda begins and her clothing ends, but I'll have to sort through the bulk of it first and find a suitable subset for training.