Offline composition: it hears the whole song first
Arcus does not rely on a microphone tracking sound in real time. The AI studies the entire song offline, before it plays, and composes a timed choreography across the whole track.
- Analyze — The whole song is read: beats, sections, build-ups, drops and the energy curve over time.
- Understand — The track's structure is mapped: where tension gathers, where the bar goes quiet just before a drop.
- Compose — A timed, anticipatory show is built from a palette of named water and light animations; it starts ahead of the drop and builds to the peak.
- Validate — Every move is tested in a real-physics 3D simulator, a digital twin of the fountain; only moves that are physically possible on the pool survive.
Three approaches: manual DMX, sound-reactive, Arcus AI
A musical-fountain show is produced in one of three ways. The difference is clearest in the table below.
| Criterion | Manual DMX timeline | Sound-reactive (envelope follower) | Arcus AI choreography |
|---|---|---|---|
| Time to make a 3-minute show | 2-4 hours to a week | Instant but crude | Minutes |
| Anticipates the music? | Yes, if the programmer designs it by hand | No, reacts a beat late | Yes, builds to the drop in advance |
| Needs a specialist? | Yes, a DMX specialist | No | No |
| Consistency / repeatability | Depends on the programmer | Every loud moment looks the same | Designed and repeatable |
| Self-serve online? | No | Partial | Yes, upload a song, get a show |
| Commands patched by hand | 1,500-6,000 | 0 (none authored) | 0 (the AI authors them) |
| Structure-aware (build-ups/drops)? | Yes, manually | No | Yes, sections/build-ups/drops |
Why sound-reactive is not choreography
A sound-reactive system is an envelope follower: it takes the loudness of the audio and maps it to jet height. It is like wiring a VU meter to a pump.
- It reacts a beat late; it follows what already happened and cannot know what is coming.
- It has no structure; every loud moment in the song looks the same.
- It does not know the drop is coming, so it cannot build to it.
Put simply: sound-reactive is a microphone wired to a pump. Arcus is a choreographer who has already heard the song.
The animation palette and the digital twin
Arcus does not draw moves pixel by pixel from scratch; it uses a palette of named water and light animations and times them to the structure of the music.
Every show is validated against a digital twin of the real hardware. A pool is made of end-units on buffer lines (up to 30 units each) under a single Main controller. Each end-unit is a jet: a motor+pump driven by a VFD (a Driver) and/or an RGB light. An optional Raspberry Pi gateway adds internet, audio and remote access.
A typical system is 150 end-units: 75 Drivers + 75 RGB across about 10 buffers. The simulator knows this topology, so the composed show runs on the real pool exactly as designed.
No specialist, no DMX timeline
In manual programming, a specialist patches 1,500-6,000 individual commands onto a DMX timeline for a single 3-minute song, and that takes 2-4 hours to a week.
Arcus removes all of it: you upload the song and the AI writes the show in minutes. The whole flow is self-serve through a browser design studio and a live 3D simulator, with no specialist and no DMX software. Arcus is made in Turkey by İkaz Elektrik and offered in Turkish, English and Arabic.