I am not so sure about that. The real hurdle at this point seems to be getting something like Dall-E to produce multiple angles and poses of a scene while maintaining characteristics of the elements in the scene. At that point it is an interpolation problem. I admit I am not an expert on these, but these do not seem insurmountable. I would not be surprised if we see something that can make very convincing short videos in less than a year. At that point, since a "short video" is really a scene, we are basically at film stage.

But silent films, no? Or where are things at as far as generated audio that syncs to the video? (sound effects, dialogue, music - each of which will have its own challenges)

Conceptually -- if you're generating the video from a script, you can feed the script and video into a foley and voice generator that recognizes "door slammed" and inserts the kind of sound associated with a door slam and that recognizes face changes and generates speech audio keyed to match. Those tasks will be solved independently on the way to long-form video itself getting sorted because you can exercise them with non-AI video along the way.

I'd be shocked if somebody hasn't already made some good demos of the foley and lip sync stuff.

This can be done with img2img and Controlnet on Stable Diffusion. But still a lot of manual work involved at least in my setup.

I doubt AI (by itself) will be generating movies like Disney movies any time ever, tbh.

But I don't doubt that it'll generate enough garbage appealing to enough people that it will account for Disney-like watch time.