I wanted to see what happens when you stop prompting AI video like an image — and start directing it like a camera.

This is COLOSSUS, my submission for the CapCut Seedance 2.5 Video Challenge.

The experiment was simple:

30 seconds. One continuous take. No cuts.

I designed the sequence like a next-generation AAA gameplay moment: a dragon and rider dive from the mountains into a canyon, race above a river, break through a waterfall, enter a ruined civilization, encounter something impossibly large, and transition from aerial pursuit into third-person action — all through one continuous camera movement.

The biggest challenge wasn't generating individual beautiful moments.

It was continuity.

Instead of describing the film in one long paragraph, I structured the prompt around timestamps, physical camera movement, character choreography, environmental geography and explicit continuity constraints.

One principle guided the entire sequence:

The camera must always have a physical reason to move.

That changed the way I approached the generation.

Rather than treating each environment as a new scene, I treated the entire 30 seconds as one connected space the camera had to physically travel through.

The result is probably my favorite experiment with generative video so far — especially the transition from dragon flight to freefall and then into a third-person gameplay camera.

And, of course, what is waiting beyond the mountains.

Created in CapCut Video Studio with Seedance 2.5 for the Contra × CapCut challenge.

#contraseedancechallenge #CapCutCPP #Seedance25 #AIFilmmaking #AIVideo #GenerativeAI #CreativeAI #OneTake