I don’t have time to make videos.
Storyboard, script, revise, record, edit. I’m building Fantasy Joes solo, and doing all that by hand was never going to happen. But I want short-form video to work before the season, so I built a system to make them instead.
Here’s the most recent one:
I’ve got a dozen or so up now, and I’m not happy with any of them. They still feel a bit like AI slop, and they’re just not good enough yet.
The format is the product
The video that fits best is basically the product itself: two players, side by side, and you pick the one you’d rather draft. I’ve tried other angles, but this is the one that works, because that choice is exactly what Fantasy Joes is. And I’ve got real data on how people make those calls.
What the system does
The whole thing runs as a pipeline. It pulls the real data, writes the script, draws the art in the app’s comic style, generates the voice, lays out every frame, and renders the final video. It checks its own work, too. Every number on screen gets verified against the real stat before it makes the cut, and a critic reviews the finished video and flags anything weak before it goes out. On the last one, the critic caught three things worth fixing, and they were fixed and re-rendered before it ever got to me.
That’s the part I’m excited about. The system is built out and tested enough now that it can turn out videos at the current quality with almost no input from me. The videos mostly make themselves. My time goes into making the machine better. After each one, I go back and refine the pipeline itself.
Where this stands
The most recent videos are the best I’ve made, and they’re getting a decent bit more views than the early ones. Still tiny numbers, but it’s moving, and there’s a bit more interaction. They’re not good yet, but it feels like it’s finally getting somewhere.
By the time the season’s here, I want a short-form video engine that runs on its own, puts out a lot of videos, and keeps getting better while I build everything else.