Satisfying video generated from a song: the bounce that lands exactly

So-called satisfying videos all look alike: a ball, a course, some colours. Some hold attention for a full minute; others get swiped away in two seconds. The difference lies neither in the scenery nor in the render quality.

It lies in what happens at the moment of contact.

What you hear without listening

When a ball touches a platform and a sound arrives at the same instant, the brain binds the two. That is a reflex, not an opinion: a simultaneous audiovisual event registers more strongly than the sum of its two halves. Nobody thinks "that is synchronised" — they simply find that it works.

When contact and sound are a few hundredths of a second apart, the binding breaks. No one could say why, but attention drops — and on a feed scrolled with a thumb, attention that drops is a finger that moves on.

Why most videos miss that moment

The usual method is to simulate a ball in a physics engine and hope it lands on the beat. That cannot work: the instant of contact depends on height, slope, gravity and the previous bounce — on an accumulation of continuous quantities. The beat, by contrast, is a grid. Nothing forces the first onto the second.

So it gets patched by hand: move a platform, slow a passage down, cut in the edit. That survives three impacts. Rarely thirty.

What this engine does

It takes the problem the other way round. The contact instants are given by the track — they are its onsets, detected per frequency band. The geometry of the course is then derived: each platform is placed exactly where it must be for the ball to arrive at the intended instant.

The trajectory is not simulated and then corrected. It is solved, in closed form, from the imposed instants. That is why the synchronisation is not approximate: it is exact by construction, and it stays exact for the whole track.

What this video shows

A fifteen-second extract. Every contact coincides with an onset — kick, snare, hi-hat. Nothing was moved by hand, nothing was slowed down to land on time.

Watch one impact, then let it run. Repetition is what convinces: thirty coincidences in a row do not happen by chance.

Getting the same thing

Drop a track, pick a look, collect the vertical video at 1080x1920 and 60 frames per second. The format is native: nothing to recrop before publishing.