I trained edm2 on my own dataset and generated 20 images. I found 15 among them are of good quality, but the remaining 5 are quite bad and blurred. I used the edm2-xs model size and default setting provided in this repo. Is the bad quality resulted from small model size or other reasons?