Description:
I’ve been experimenting with MiniMax’s H3 Max Director model. Picture this: you upload a song, and the model is asked to interpret it through dance. As the performance unfolds, you can ask for more dancers, introduce a chair, make it rain, or even have the dancers move as if they were on the moon. The video is generated as you watch. Nothing is rendered in advance. Your directions influence upcoming footage.
It’s far from perfect: glitches, strange movements, inconsistent quality, and instructions that don’t always land. But directing through suggestions is fascinating. You don’t know exactly how they’ll be interpreted, and that unpredictability feels like part of the medium.
Somewhere between watching, playing, and directing.