I thought difussion algorithms were essentially Markov chains applied to images instead of text, so they would be an LLM with the language being images. But honestly never cared to look to deep into it so I'll take your word for it that they're different.