Generating sprites on the fly would be very taxing, since it would require either the dev to have another cloud instance for Image gen that everyone could use the current tokens for, or have you run a local image gen engine in attention to the local LLM. It would be cool to get custom CGs like that, but I don't think it's very viable. Keep in mind that the sprites in-game are NOT just raw generated, but have a lot of Img2img touch ups done, or at least I believe that's what Three Eyes said. So raw image gen of the sprites wouldn't look the exact same/nearly as good.
Chat with Monsters wouldn't work because of the way the LLM context is set up. It MIGHT in theory, but it would be extremely messy and no where near as good as the current Agent setup with the townsfolk.
I'd love to see it tbh, but there's a lot of limitation that prevent it from happening. I wouldn't mind seeing the MC able to move around in chat like NPCs can, though, and I think it's do-able?