A lot of games give you answer options with most of the time, one of them is the stupid answer, that makes the protagonist at least a troll or an idiot. But i guess they doesn't mean that.
I'm not sure if we really need realistic graphic. The advantage of most jrpg's is, that we usualy have anime style art. Sure the model get here as well more and more complex, but they doesn't have to be that detailed as realistic graphic and maybe we don't want that as well. About the silent protagonist...well i guess i prefer the nto silent one. But if we have one, giving them some kind of character, like joker (persona 5) is something i prefer.
With AI being shoved everywhere, I expect to just be able to talk to the NPCs with my own voice at some point. I already do this with silent protagonists in games, but in the future they will actually hear and respond to my bullshit.
For what it's worth, the game devs and LLM experts I've spoken to on the matter don't seem to think that's coming without another big advancement in AI tech. Some games will try it, but similar to the visual phenomenon, there will be an "uncanny valley" with dialogue that won't be crossed anytime soon.
It's definitely plateaued, Stable Diffusion 3 came out and people just didn't stop using 1.5 / xl because 3 was just not any better.
I think we will get better workflows and better ways to handle the ai we already have, using control nets and other control functions / multiple ais etc