GX Aura IllustriousXL LoRA
This one's gonna have to be remade. I've got a double hurdle of having two outfits in one lora on top of one of them involving print clothing. Despite me adding tokens to exclude, it seems the plushie she holds in a good number of the images I have of her anniversary outfit had been burned in as being the design on the shirt, so it doesn't come out correctly (at least, that's my best guess). The base outfit is kinda okay but not super accurate.
I have later steps and they seem to start blending the outfits into each other so I'll have to remove any shared clothing tags or separate them into two datasets with triggers. Whatever, at least the hairstyles and face are good
SLAG LOVE
https://mega.nz/folder/dfBQDDBJ#3RLMrU3gZmO6uj167o-YZg/file/oCQjCTADI also tried to make a new Dr Nova lora for Illustrious. I threw in some of the 3d screenshots used for the 3d lora to help with accuracy and get some more data of her without the labcoat on (since it's over represented to the point of burning in a bit), but it turns out—and I should have guessed this from the way the 3d lora behaved—3d screenshots and 2d illustrations, despite how well a human brain can interpret them as being the same art style, are interpreted as completely different by neural networks. The resulting lora, even at 2250 steps, acted as if I was training two different characters and styles in one. Neither looked right, without the 3d tag it looked worse (despite drawings and live2d shots making up two thirds of the dataset). Will have to try something else.
