The movement writes the art, not the prompt.
My first version described a scene: a girl, a street in Japan, doodles on the walls. It looked good. But it only worked for that one clip.
So I threw it out. The new prompt never says who is in the video or what they're doing. It only gives rules. A body part that moves leaves a mark along its path. Bigger movement, bolder marks. No movement, no marks.
That's why it works on a slow walk, a breakdancer, or someone waving in a parking garage.