I think the fact those diffusion models are smaller and compute efficient than gigantic GPT models are in fact make them easier to use and distribute.

BLOOM is out there, but not that many individuals with have like 8 3090 to host them, and the inference is still incredibly slow nevertheless