Hi, thanks for your awesome work. I followed your instruction to finetune Stable Diffusion on the pokemon dataset. There are no problems in my training process, but my model is not as good as yours. It can generate image in pokemon style but seems to forget old information such as Donald trump or Obama. I tried your model on HF, it did not forget Obama and can generate better image without weird detail. I tried to modify the warm up steps to 10000, decrease Lr a little bit. The result did get better but still not as good as yours. So I wonder what is your configuration? Thank you.
Hi, thanks for your awesome work. I followed your instruction to finetune Stable Diffusion on the pokemon dataset. There are no problems in my training process, but my model is not as good as yours. It can generate image in pokemon style but seems to forget old information such as Donald trump or Obama. I tried your model on HF, it did not forget Obama and can generate better image without weird detail. I tried to modify the warm up steps to 10000, decrease Lr a little bit. The result did get better but still not as good as yours. So I wonder what is your configuration? Thank you.