One could definitely build a sort of RPGMaker framework engine kind of thing on traditional procgen methods that builds a bunch of prompts to get assets from text and image generators while the engine itself provides the structure and all the mechanical bits. Hell, I could probably have the concept working in a month or two of work, depending on how low the bar for "it makes a game" is.

It would just suck horribly and make the actual worst nonsense you've ever seen. Like just as I can see how all the parts to make this would fit together I can also see how everywhere the generators are involved will just straight up fail and produce absolute random garbage because that's what they do, they produce nonsense that at best only looks good in isolation and to do any better than that they have to both be actively shackled to a very narrow task with tools that require human curation and design tailored to the situation and even then their outputs still have to be carefully curated and edited by hand afterwards.