Try size: 768 x 1024 or above
Weights: 1.0
Clip skip: 2
Training material from bing
Pictures from kantanmix_v16LCMTurbo
Note: Sometimes the "convex mirror" will be too big and even float in the air, but the probability is not very high, so it has little impact. Because ILXL has specially trained the "traffic mirror" (danbooru tags), so I specially added it this time.
