Style LoRA that makes everything flat and cartoony.
Activation word is FlatPop
Trained on 697 anime girl pictures. Synthetic dataset available for download too.
Illustrious works great. Pony likes to cut off people's heads. SD1.5 varies wildly depending on your checkpoint.
I have 2 Klein versions, "A" is done, "B" is 3/4 the way through training. It's style's a little closer, but it has a hard time making genitals.
Will retrain ZIT with new settings soon
I made a great tutorial on how to prepare images
Description
Extremely experimental first attempt at a Z Image Turbo version of this model.
Will be redoing to make it work properly.
FAQ
Comments (3)
do you need a special configuration to use LORA with z-imageturbo? Despite downloading the correct ones, it's like they don't apply at all, nothing happens really, it's the usual result of base model
Right now it only works in ComfyUI and not Automatic1111 or other variants.
Browse Templates > "z-image-turbo text to image"
And add a "LoraLoaderModelOnly" node right after the "Load Diffusion Model" node.
Make sure to still mention the trigger word at the beginning.
Z Image Turbo doesn't use the booru tag prompts like stable diffusion, it expects full sentences.
TRAINING a LoRA for it requires an entirely custom weird setup.
@fudruckers i'm using forge neo, guess i'm fucked for now, thanks anyway!