The LORA is designed to generate CHIBI style images, and the backgrounds generated by the model are also more lovely and flat traced.
Since sdxl is trained with 1024*1024 resolution images compared to sd1.5's 512*512, I chose to adjust the dim to 256 for training, so the size of this lora is 1.7G.
Please use with SDXL1.0 or counterfeitxl.
Recommended lora weight 1
0.8 or less lora's influence will be reduced a lot.
