This will get better as NN/AI chips become the norm in gaming. Compression gains, on the fly generation of textures, voice generation when needed, etc.
I envision a future dev using rough shitty textures to conceptualise a game, and then an NN to bring it to life during runtime.
You might even be able to load your own NN interpreter to make the world more cartoony, or change the intended setting entirely, or unlock the nsfw filter on the vanilla interpreter.