Diffusion versions had been released a yr afterwards by researchers at Stanford University along with the College of California at Berkeley. By iteratively refining their output, these models figure out how to make new knowledge samples that resemble samples in a very training dataset, and happen to be employed to make real looking-looking images.