Can text-to-video AI models really work long term?

08 Tem 21:38

Text-to-video is wild - we’re not there yet, but the leaps from last year to this are insane. I remember testing Runway’s early model last year and the results were... let’s just say if you blinked you missed the faceplant. But now with models like Sora or Pika Labs, basic physics, camera movements and even 10-second clips with semi-consistent characters are suddenly in the "good enough for internal pitches" territory. Even in my workflow I’ve started using these for rough storyboards or placeholder assets when client feedback changes faster than my artists can iterate. Yeah, the hands still look like abstract art in most cases and temporal consistency is the first thing to go out the window, but for quick internal mockups or social content drafts? Saves *days* of waiting on animation revisions. So long-term? Not replacing Pixar tomorrow, but giving traditional pipelines a run for their money in the "good-enough, fast enough" bracket. The data hunger is real - these models swallow terabytes of video to learn physics from - but the trajectory is undeniable. Five years from now? I wouldn’t be surprised if we’re supplementing or even replacing early production phases entirely.

08 Tem 23:00

I’ve seen a few demos of these tools and yeah, the physics is still wonky—like a character’s arm stretching out randomly mid-scene. But when it nails the small stuff, it’s wild how quick you can rough out a storyboard. For short clips or ads, it’s already a time-saver; full feature films might need humans calling the shots for a while though.