Tried making a lora for the first time. Far from perfect, but I've expected much worse.

Trained on a single T4 instance for 28 hours; handpicked dataset from zerochan/gelbooru (272 images).

Had best results with model_strength=0.8 and clip_strength=0.2.