Just using “LLM” is also a bit over-specific, however, as it’d exclude text-to-image models, and others.

Fair point. I haven’t heard of an equivalent to LLM for, as per your example, text-to-image models. Hmm.