Cutting to a beat automatically, and where the machine hands the edit back

A music-led edit involves finding the beat, marking it, and then deciding which of those marks become cuts. The first part is arithmetic on an audio waveform. The second is the edit.

People either do both by hand — tapping along, placing markers one at a time — or try to automate both and end up with a cut on every beat, which is exhausting to watch and reads as a demo rather than a piece.

The productive split is to automate the marking completely and leave every decision about cutting to the editor.

Beat detection produces a list of times in seconds where a beat occurs. Several libraries do this well and the output is the same shape regardless of which you use.

Two things to check on the result. First, whether it found the tempo or half or double it — a common failure, and obvious once you look at the spacing. Second, whether it drifts on tracks that are not strictly metronomic; live playing will wander and the detected grid will not follow.

Listen to a click generated from the detected times before using them. Thirty seconds, and it catches both problems immediately.

Beats arrive in seconds; timelines work in frames. The conversion needs the project's actual frame rate, and rounding matters — a beat landing half a frame either side of a cut is the difference between tight and slightly loose.

Round to the nearest frame rather than truncating, and be careful with non-integer frame rates, where accumulated error over a three-minute track will visibly drift by the end.

Place the markers on the timeline with the beat number in the name. Numbered markers let you work in bars — every fourth, every eighth — which is how you will actually use them.

A cut on every beat is relentless. What works is cutting on a subset that reflects the music's structure — the downbeat of each bar, or every second bar, with denser cutting reserved for sections that build.

Numbered markers make this straightforward: work on every fourth or eighth marker, and drop to finer subdivisions only where the section warrants it.

The other technique worth knowing is deliberately not cutting on a beat the viewer expects. A held shot through an anticipated cut point creates tension precisely because the grid established the expectation.

It is tempting to go one step further and have a script lay clips at every marker. The result is technically synchronised and has no relationship to content — a shot ends because the beat arrived, not because it was finished.

Where this is acceptable is a fast montage with interchangeable shots. Everywhere else it produces a sequence that feels mechanical in a way viewers notice without identifying.

Stop the automation at markers. The marker grid is genuinely useful and it took the tedious part; what happens between the markers is the reason there is an editor.

Write the detected times to a file in the project folder alongside the track. Re-cuts, alternative versions and different durations all use the same grid.

Include the frame rate the conversion assumed. A beat map is only correct for one frame rate, and applying one from a twenty-five frame project to a twenty-four frame version produces a drift that is maddening to diagnose.

With the map saved, producing a shorter cut of the same piece starts from a grid rather than from the audio again.