Vaib

The problem

Programming stage lighting is slow, expensive, and gatekept by a small number of specialists. A touring show can lose days to it. A smaller venue simply goes without, and the show is worse for it.

Most software in this space still thinks in DMX values, channels and cues — the vocabulary of the machine, not of the person trying to light a song.

What makes it hard

Generating something that looks like a lighting design is easy. Generating one that the rig in the room can physically execute is not.

Every venue has a different set of fixtures. Each fixture can do particular things: how far it moves, how fast, what colours it can actually reach, whether it can strobe. A design that ignores those limits is worthless — worse than worthless, because someone has to notice and fix it before the doors open.

So the hard part is not generation. It is constraint: the output has to be bounded by a live model of the specific hardware present, not by a generic notion of “stage lighting”.

What we did

Vaib learns the properties of the actual fixtures in a rig, and combines that with training on hundreds of concerts to produce a lighting design that suits the music — and that the rig can perform.

Testing happens where it matters. Lighting only reveals itself in a real room with real acoustics, real sightlines and a real audience.

Where it stands

In testing at live events. It is the clearest example of how we approach AI: the model is the easy part, and the value is in bounding it with reality.