So having tested this, I'm pretty sure the issue is your training data, which also appear to be an insufficiently diverse dataset. I'm guessing you used synthetic data (maybe something produced with Pony models?). This could be where the shiny skin is coming from.

With regard to dataset diversity, this model appears to have been pretty much exclusively trained on thin, young, Latino men. It is nearly incapable of producing men of other races, ages, or body types, and ignores prompting in this regard, even with guidance cranked way up.