with this "precision" setting, should you always try to match it to what the model's precision is?
(you can set precision with text encoder, vae, and the main model)
if I set it to fp32 it takes longer and I get different outputs but I'm having trouble figuring out if the quality is better.