V2

• Increased accuracy for corrupted forms and other low representation outfits

• Increased flexibility

• Partial support for outfits on other characters. Possibly more in future versions

V0 Notes:

Quick and dirty draft version of Exs-Tia Chevalier LoRA. I have two possible dataset branches to test next, whenever I get the buzz for training. Note that the trigger words will change in future versions. I'm not sure whether I am going to try and include as many outfits as my Marina LoRA within this one in v2 onwards or just strengthen a few outfits. I think I will be pruning at least a bit. I used both the newer and older artstyle in the dataset, which muddied the results.

May have been an issue with my setup while I was playing around with the LoRA, but I absolute nonsense outputs a few times. slaughter-princess-outfit does not work reliably at all. I'm not including it on the side in the trigger words for that reason.

LoRA was trained to be incredibly flexible... but really only the dancer outfit and chevalier outfit triggers will be reliable. Make sure to add supplemental tags for dancer outfit, and it should work. Mostly just chevalier form works well in this one. You can probably "build" most of her outfits by specifying them. I put in a lot of supplemental training information for more unique generations at the cost of the outfits as an experiment.