Can one GPU make a whole show?

This entire scene started on one computer in a spare room. The outline came from a local language model. The voice came from an open model running on the same GPU. Five images became a moving sequence, then the render pipeline added timing, sound, captions, and a real experiment ID. No stock footage. No mystery invoice. The useful question is not whether AI made it. The useful question is whether the result held your attention. What part should we open up next: the voice, the images, or the automation?