It's hilarious to read back in this thread and realize that just a month ago, we had pundits in this very thread nodding sagely and explaining to us that making a movie was thousands, perhaps millions of times harder than chatting or making images and well beyond what AI would be able to do in the foreseeable future.
The post being linked demonstrates incredible difficulty with stability for
static scenes, with the shift having nearly total change of the "planets" arriving between frames, with a total of thirty seconds shown. Same problem crops up in
Disturbed's Bad Man video, and every other AI animation I've seen, they are
all hellishly unstable to the point of being nearly totally worthless for producing conventional content.
Except the AI Seinfield, but that goes to
extreme lengths to reduce the complexity to the point its barely worth considering for end-use
and still has visible consistency errors like a "box" with an edge that switches from concave to convex mid-scene. Provided that the "clipping" behavior and certain other artifacts aren't the result of using "dolls" to completely bypass a lot of the difficulty instead of direct AI generation of footage.
I lived through the 80s. I learned the hard way not to say "It's beyond possibility that computers can...."
The difference is that we're running out of curve on silicone and the current methodology has little potential for optimizing the problem
at the scale worried about, so continued progression is the domain of wholly new technology. Maybe quantum computing will leave the "coming soon!" it's been in for over a decade in the next five years, maybe dedicated architecture magic will bridge the gap with existing manufacturing like it warped cryptocurrency, maybe a new AI methodology will bypass the technical challenges.
But because we
physically can't brute-force it by throwing more transistors at it, it is uncertain. This is "Fusion is 10 years away!" for the last 50 years thinking, that because visible progress is being made it's
obviously close to being sold to the end-user. Maybe it's 5 years, maybe it's 10, maybe the technical challenges will keep cropping up for the next 50 years like they have for fusion.