[skp1995]

I am missing the link to the thread, but diffusion models also give a very consistent output when prompted with `IMG_{number}` part of the reason could be the training data distribution