A Week of Experimenting with Video Models

Seven days, five video models, one GPU that sounded like a jet engine. Notes on what worked, what didn't and what surprised me.

I gave myself one week and one rule: build one tiny thing per model, every day, and write down what broke.

The surprise winner wasn't the biggest model — it was the one with the most predictable motion. Consistency beats spectacle when you're building tools.