THE TOOLCHAIN
Notch: real-time generated effects that react during the show
- ROLE
- Real-time rendering
- TYPICAL USE
- Reactive content
- RUNS
- Show machine · GPU
Computed at showtime, not rendered last week
Playback content is decided in advance. Notch content is computed while the show is happening, which is the whole argument for it, and also the whole risk.
Notch is published by Notch; the program itself lives at www.notch.one, and what follows is what it is for rather than what it ships.
Notch answers a question that playback cannot: what if the picture has to respond to something that has not happened yet? Its content is real-time rendered during the performance, which makes it capable of reacting, and makes it a live system with all that implies.
The case for generative
Three situations make generative content the right answer. When the source of truth is live (audio, tracking, an operator’s hands) nothing pre-rendered can follow it. When the material would be impractical to store as video at the resolution required. And when the same block has to behave differently across a run without becoming a different asset.
Outside those, playback is usually the better engineering choice, and experienced designers reach for it first.
What it costs on the show machine
Real-time rendering happens on the same hardware that is decoding video, driving outputs and running the operator’s interface. Live rendering is not free capacity; it is capacity taken from everything else.
The second cost is a different reliability profile. A video file either plays or does not, and you know which during rehearsal. A generative block behaves differently depending on its inputs, which means it can be fine for three weeks and then meet an input nobody tested. Rigs that lean on it budget hardware headroom and test at worst case rather than typical.
Playback content is decided in advance and is therefore known to be acceptable before the doors open. This is not, and that is the entire trade: a picture that can respond to what is actually happening in the room is also a picture nobody has approved.
The discipline that makes it survivable is bounding the parameter space. A generative system with an unbounded input can produce a state nobody has seen; the same system with clamped ranges produces variety inside a set of outcomes that have all been looked at.
Where it sits in the rig
Almost always as a source rather than as the output stage. The block is loaded into the playback environment and composited like any other layer, so the operator’s workflow does not change, some cells are files and some are computed, and the grid does not care.
It appears most often on volumetric work, where content has to be sampled per layer rather than drawn per layer; see LED and volumetric and the volumetric array.
What it costs on the show machine
Real-time rendering competes for the same GPU that is driving the outputs, and unlike playback its cost is not constant. A scene that runs comfortably in rehearsal can drop frames when a parameter reaches a value it never reached in rehearsal, and that is exactly the situation the technique exists to allow.
The mitigations are unglamorous and known: cap what can be spawned, test at the extremes rather than the middle, and keep a playback fallback for the sections that must not fail.
When it is the right answer
When the picture has to react. A show that runs to timecode with a fixed set list rarely needs it; a set that breathes, a tracked scenic object, or an audio-reactive section does. The operator is usually the one holding it, and the systems tool is usually what feeds it control data.
What it is not is a substitute for authored material; see content creation for the slow half of the job, and the toolchain for how the two divide.
It is also worth being clear about what it does not remove. Real-time content still has to be designed, and a generative system with nothing to say produces novelty rather than a picture. The technique changes when the frame is decided, not whether somebody decided it.
Questions
the things people ask about thisWhat does Notch do that a video file cannot?
React. Notch content is real-time rendered during the show, so it can respond to audio, to timecode, to operator input or to tracking data. A video file is a decision made weeks ago; a generative block is a decision made at the moment it plays.
When is generative content the right choice?
When the picture must respond to something unknown in advance (an improvising performer, a live audio feed, a moving tracked object) or when the material would be impractical to render and store as video at the required resolution.
What does it cost to run live?
GPU capacity on the show machine, and reliability attention. Live rendering competes for the same hardware as playback, and unlike a video file it can fail in ways that depend on input. Rigs that use it heavily specify the machine around it and test under worst case.
Does Notch replace playback content?
No. Most shows are mostly playback, with generative blocks where reaction is needed. Playback is predictable, cheap to run and known-good; real-time earns its place where predictability is exactly what you do not want.
How does it fit with the VJ software?
Usually as a source. A Notch block is loaded into the playback environment and treated as another clip that happens to be computed rather than read from disk, so Resolume composites it alongside everything else.