A lora trined on images of "bad taxidermy". It turned out fun as expected :D

V1: trained on just 30 images, it's fun :D

V2: trained on a larger dataset (about 100 images). I'm not sure it's significantly better than V1 but in theory it should be more flexible