SD(1.0-0.8)
SDXL/PONY see for yourself.
Training in 1024, larger image database, smaller file size, slightly corrected the problem with double-triple eyebrows.
smug3(sdxl).safetensors