Diffusion models' representations show alignment comparable to smaller datasets, with text conditioning enhancing alignment.
