What happened: Text-to-video models went from shaky 4-second clips to minute-plus scenes with consistent characters, backgrounds, and basic camera control — a jump most people noticed sometime in the last year.

Why it matters: Longer, more coherent output opens real production use cases — previsualization, ad drafts, synthetic training data for other models — beyond the novelty demos that defined the category early on.

How it works, roughly

Not frame-by-frame image generation stitched together — that's what caused the flicker and inconsistency in early models.…