TikTok videos synced to the music: the example and the method

There are plenty of videos "with music underneath". There are very few where the movement lands exactly on the sound. The difference is not something you explain, it is something you see — and that is precisely why it works on a feed people scroll with their thumb.

What this video shows

A ball falls. It hits a platform, bounces, hits another. Every contact coincides with an onset in the track: the kick, the snare, the hi-hat. Nothing is edited by hand, nothing is slowed down to land on time — the trajectory was computed from the track, then rendered in 3D.

The format is natively vertical, 1080×1920 at 60 frames per second. That is what TikTok, Reels and Shorts ask for, and it avoids the crop that eats the top and bottom of a video shot in landscape.

Why exact sync holds the eye

Human vision is an event detector. It does not "watch" an image continuously: it picks up changes in motion and alignments. When a visual impact coincides with an audio impact, the two reinforce each other, and the brain records one event rather than two approximate ones.

Three concrete consequences on a video platform:

• The first second decides. A clean impact right at the open gives a reason

not to scroll. It is the only moment where you genuinely win attention.

• The loop becomes natural. A video whose last impact lands on the beat loops

seamlessly. Replays count towards total watch time.

• Approximation costs you. Even a small offset produces a vague discomfort

the viewer never identifies but does feel. They do not think "that is out of sync"; they scroll.

What "exact" means here, in milliseconds

The word deserves a number, otherwise it is advertising copy.

A video at 60 frames per second cannot show a contact between two frames. Every beat time found in the track is therefore snapped to the nearest frame. The maximum error is half a frame — 8.3 milliseconds.

The threshold beyond which eye and ear start to perceive a lag between a sound and a visual impact sits around 20 milliseconds. So we work with better than a factor of two in hand, and that margin depends on no setting.

More importantly, the error does not accumulate: every beat is snapped independently, against the clock of the track. Bounce number four hundred is as accurate as the first. That is what separates computation from physics simulation, where the error of one bounce becomes the starting error of the next and ends up producing tens of frames of drift across a full track.

Beat strength drives the height

Perfect sync that is monotonous is still monotonous. What makes a video musical rather than metronomic is that the trajectory responds to intensity: a heavy kick throws the ball high, a hi-hat barely nudges it.

The analysis picks up onsets band by frequency band — lows, mids, highs — instead of merging them into a single stream. That separation is what makes it possible to tell the hits apart and give them different amplitudes.

Format matters as much as content

Three points with nothing spectacular about them that nonetheless decide the fate of a post:

1. Natively vertical. 1080×1920, not a landscape video cropped down. A

cropped video loses its composition, and it shows immediately.

1. 60 frames per second. The fast motion of a fall stays readable; at 30, the

impacts blur and lose their edge.

1. Sound is the argument. A video whose whole point is sync cannot be judged

in silence. That is also why this page does not start playing on its own: browsers block unsolicited audio, and a silent demo demonstrates nothing.

What it takes to reproduce this

An audio file. That is all.

The engine picks up the onsets of the track, turns them into a beat map, solves the trajectory backwards from those instants, builds the 3D scene and renders the video. No editing software, no graphics card on your side.

Two properties follow, and both are worth knowing before you start:

• The same track gives back exactly the same video. No unseeded randomness,

no physics iteration. For iterating on an edit that changes everything: you are not starting from a fresh variant on every render.

• Length is bounded by the plan, because a minute of video costs real minutes

of graphics card. The free plan goes up to 30 seconds, the Creator plan up to 90. The TikTok monetization page explains why that length is not a detail when you are aiming at the rewards programmes.

You remain responsible for the rights to the music you upload: a commercial track stays subject to the rules of whichever platform you post on.