Live reactive visuals: Audio reactive, Beat reactive, Frequency reactive and generative visuals.

There is a huge difference between a clip that flickers when the volume rises and a scene that breathes with the music. The first impresses on the first night; the second still works in the fourth hour, when the venue technician has lowered the overall level and the dancefloor’s energy has changed.

This guide explains the four reactivity modes found in any live visuals system—Audio reactive, Beat reactive, Frequency reactive and Generative visuals—the data each needs, how to prepare them and how to combine them without making the audience dizzy.

Four ways to respond to music

All modes share the same architecture: a data input, a process that converts it into useful values and an output that moves a visual parameter. The input is what changes.

• Audio reactive: works with the signal’s amplitude over time. It is calculated using the envelope (RMS for average energy, peak for peaks) and produces a smooth value that rises and falls with the volume.
• Frequency reactive: breaks the signal down into frequencies using an FFT and delivers separate bands, so bass, mids and highs can control different things.
• Beat reactive: looks at the clock rather than the volume: it uses BPM, beat position or detected hits to mark the precise moment of the beat.
• Generative visuals: uses the other modes as parameters, but an algorithm calculates the content on every frame: there is no repeating file.

Initially, the choice is about available data rather than aesthetics. If your system only captures audio, you have the first two; network or Link sync opens up beat-based control, and a GPU budget opens up generative content.

Audio reactive and frequency reactive

Audio reactive is the entry point. In any software with audio analysis, the signal comes through an input (capture from the mixer, interface or network) and produces an envelope normalised to a working range, usually 0-1. That value modulates brightness, scale, displacement or particle density. Its strength is that it always works; its weakness is that it responds to the overall volume: if the technician lowers the master, the scene falls asleep.

Frequency reactive addresses this by separating the spectrum. The FFT divides the signal into bands, and each band becomes a control channel: energy at 20-120 Hz can deform a mesh, 300-3000 Hz can modulate colour, and above 5 kHz can produce rapid flashes. The trick lies in two settings: the number of bands (few are stable; many are noisy) and smoothing, a filter that stops the value jumping around on every frame.

Practical preparation: first measure the input at a healthy level without peaks, then normalise the control range so the value reaches its maximum only during strong moments, and finally apply smoothing. Without that chain, the visual moves with the background noise instead of the music.

Beat reactive and tempo sync

Beat reactive changes the question: how loud it is becomes which beat we are on. The three usual sources are BPM and beat position sent by DJ software, a shared clock such as Ableton Link, or LTC/MTC timecode. With any of these, the scene receives beats and bars and marks the frame on every pulse.

It is the most reliable mode for choreographed pieces, because the hit lands the same way with loud or quiet music, and the cleanest for structure: you can program the scene to tremble in eighth notes during the build and return to pulses at the drop. It also handles variable-tempo tracks and live bands least well: those need real-time beat detection or an operator making corrections.

In practice, beat reactive almost never runs alone: it serves as a trigger (the hit switches something on), while audio reactive keeps the base lit and frequency reactive gives it texture.

Generative visuals: content that draws itself

Generative visuals calculate every frame using algorithms: particle systems, noise fields, deforming geometry, shaders. They do not repeat, do not end and respond to any parameter, including those from the three other modes. That is why they are used in long pieces: a generative background can last for hours without cutting or repeating.

The cost lies in the GPU and art direction. Making it work requires a narrow parameter range (fewer than ten well-chosen controls), a measured render budget at the actual output resolution and a fixed palette; without those, the result is a screensaver. The technical reference is the guide to audio-reactive and generative visuals, which details the operator chain that performs the analysis.

Combine modes without overwhelming the audience

The combination that works at events uses layers with separate responsibilities:

1. Base layer, generative or clip: stable, with slow movement, able to run for hours.
2. Middle layer, audio reactive: brightness and scale follow the room’s overall energy.
3. Detail layer, frequency reactive: colour and texture accents tied to bass or highs.
4. Accents, beat reactive: precise flashes and cuts at selected moments.

And one rule of discipline: only one module moves each parameter. If the overall level, spectrum and beat all control brightness at once, the image becomes unreadable. The same idea applies when a show shares data with other disciplines: if the lighting console already handles the hits, the image can stick to the spectrum. That division is agreed in cue phrasing and show control.

Settings that prevent the flashing-sign effect

Before going live, four tests can save the show: a full session at the room’s actual volume (analysis changes with level); a ceiling on flashes (limit flashing frequency and avoid extreme contrasts); a degraded mode if the audio input fails (the scene must stay alive without a signal); and a frame-rate measurement at the output resolution, rather than the preview resolution. If the system drops to 20 fps, the solution is to remove layers rather than lower the volume.

With these foundations, the image stops being a decoration that reacts to shouting and becomes another part of the show, with its own data, limits and plan B.

From rehearsal to show: how to validate a mode

A reactive mode approved on a laptop is not approved yet. Real validation has four steps and uses the final equipment.

Reference signal. Before changing parameters, send a familiar track recorded at a fixed level to the analysis input, and observe which ranges move and by how much. That recording also serves as a regression test: if the scene responds differently to the same file a month later, something has changed in the chain.

Range and smoothing. Adjust the value so it reaches its maximum only at strong moments and the background stays near the bottom of the range. Then apply smoothing: too little produces a jittery jump; too much makes the scene arrive late on every hit. Check by listening: if the visual flash arrives noticeably after the kick, reduce the smoothing.

Resolution and render budget. Check the frame rate at the actual output resolution with all layers active, rather than in preview. If the system cannot keep up, cut layers or lower the render resolution before giving up the mode.

Test with a pretend audience. Play the entire show at room volume with people in front of it for ten minutes. This is when things nobody sees on the computer appear: visual fatigue, uncomfortable flashing, moments when the image clashes with the venue’s lighting.

Document the validation result: range values, smoothing, resolution, active layers and a photo of the base scene. With that, anyone on the team can reconstruct the correct state without asking the creator.

Next step

Once the image responds, the next step is to connect it to the rest of the performance: cues, phrases and coordinated triggers in Show control: cues, phrasing and VJ integration.