I kept seeing portfolio sites where an animated character reacts to you. I liked them, and I wanted one, but I didn't want the same girl at the same laptop waving hello. So I asked myself what my version of that gesture would be. The answer was the thing I do all day: framing a shot.

The idea is simple. You land on a wide frame of me, or a painted version of me, on a rooftop at sunset. As you scroll, I bring my hands together into a director's frame and the camera pushes in until you're looking at my eyes through my fingers. Then that rectangle opens into the work.

What I tried

Two keyframes, one five-second generation between them. That's the whole film. It took four versions. The first had my hands move first and the camera follow, like two separate decisions. The second lost a ring halfway through. The third gave me a sixth finger, and the fourth was clean only after I worked out that the extra finger was already in my end frame. The video model was faithfully copying my mistake.

Slow scrolling is the harshest QA there is.

At normal speed, all four versions looked fine. Scrubbed frame by frame under your thumb, every flaw sits there waiting for you. A ring that vanishes for three frames becomes impossible to unsee.

What I took from it

Describe hands finger by finger, not "five fingers". Ask for the camera and the action to start together and say what the halfway point looks like, or the model will do them one after the other. And check the still before you spend credits on the motion.

It's five seconds long, and it's the piece on this site I've looked at most closely.