MIDI Learn answers "which parameter does this stick move?" It has no way to answer "and how should it move?" The gamepad MIDI node graph is built for that second question, and it is where most of the interesting behaviour lives — the smoothing, the curve, the hold, the quantize, the thing that keeps modulating while your hands are somewhere else. The node graph exists for the gap between the input and the output.
- What it is: a transform layer between raw controller input and MIDI out. Patch nodes between the two.
- How many nodes: 79 built in — 21 shape, 15 modulator, 12 gate, 11 logic, 10 spatial, 9 music.
- Start with two nodes, not ten. Dead Zone and Smooth fix most real complaints on their own.
- It is a Pro feature, and the graph saves into the mapping so it travels with your preset.
What MIDI Learn cannot express
Learn is a binding. Stick axis goes to filter cutoff, done — and the relationship is fixed, linear and instantaneous. That is genuinely all you need a lot of the time, and the MIDI Learn guide covers doing it well.
Here is what it cannot do, and every one of these is a real complaint about using a gamepad as a controller:
- Stop a worn stick jittering a CC when your thumb is nowhere near it.
- Hold a value after you let go, instead of the spring dragging the parameter back to centre.
- Fire on a double tap but not a single one.
- Snap a sweep to notes in a key, so a stick plays something musical rather than a chromatic smear.
- Keep modulating while your hands are busy on something else entirely.
None of those are binding problems. They are all questions about the signal on its way through, which is exactly what a node graph is for.
How it is actually structured
- Input: R-trigger — analog axis
- Node: Curve — exponential
- Node: LFO — depth
- Node: Quantize — to scale
- Output: CC 74 — filter cutoff
Controller sources on one side, MIDI sinks on the other, and whatever you patch in between. Each tick the bridge reads your controller, walks the graph in topological order, applies each node in turn, and accumulates what comes out at the sinks. Cycles are refused at evaluation time rather than allowed to spin.
The nodes are grouped the way a modular rack is, so the picker is navigable rather than a flat list of seventy-nine things:
| Category | Count | What lives there |
|---|---|---|
| Shape | 21 | Dead Zone, Curve Shaper, Clamp, Smooth, Slew Limit, Range Remap, Hysteresis |
| Modulator | 15 | LFO, ADSR Envelope, Sample & Hold, Random Walk, Tap Tempo, Peak Hold |
| Gate | 12 | Toggle, Tap vs Hold, Double Click, Live Gate, Pulse, Panic |
| Logic | 11 | AND, OR, XOR, NOT, Comparator, Counter, Mux 2-to-1 and 4-to-1 |
| Spatial | 10 | Polar, XY Quadrant, Radial Sector, Vec2 Rotate, Flick Detect |
| Music | 9 | Scale Quantize, Chord Build, Transpose, Octave Shift, Velocity Curve |
Three patches worth building first
1. Kill stick drift properly
Stick X → Dead Zone → CC. A worn analogue stick does not rest at exactly centre, so it sends a trickle of CC changes while you are not touching it. A dead zone node ignores everything inside a radius of centre. This is the cheapest possible patch and it fixes the most common complaint about gamepads as controllers.
If the value is still noisy while you are moving, add a Smooth after it. The stick smoothing write-up covers which filter suits which gesture, and the deadzone and curve maths covers choosing the radius rather than guessing it.
2. Make a stick play in key
Stick Y → Smooth → Scale Quantize → Note. Smoothing first, quantizing last — in that order. Reverse them and you are quantizing jitter and then averaging the result, which produces a value that is neither smooth nor in key.
3. Modulation that runs without you
LFO → Range Remap → CC. The LFO node is free-running: it tracks wall-clock time rather than counting ticks, so its frequency stays accurate even when the bridge loop jitters. Shapes are sine, triangle, ramp up, ramp down and square, with a rate in Hz and a duty control for the square.
One detail that catches people: its output is always 0 to 1. If you want a different range, that is what the Range Remap node is for — patch it downstream rather than looking for a range control on the LFO.
Where the graph lives
In the mapping. That matters more than it sounds: the patch travels with your preset, so loading that preset in another session brings the behaviour with it instead of leaving it behind in application state. Publish the preset to the marketplace and the graph goes with it.
Is it worth the Pro licence on its own
Honestly, it depends what is stopping you. If a dead zone and a smoothing node fix your stick, that is a real fix and you will use it every session. If you were going to patch LFOs into three parameters and build logic gates, it is the most differentiated thing in the app.
If you have not tried the bridge at all yet, start with the free tier and confirm your controller is detected — every button, stick and trigger moving live on screen. That takes five minutes and tells you whether any of the rest is worth your time.
FAQ
What is the node graph in Universal Controller MIDI?
A transform layer between the raw controller input and the MIDI output. Instead of a stick axis going straight to a CC, it passes through nodes you patch together — curves, smoothing, LFOs, logic gates, scale quantizers — and the result is what your DAW receives. It is a Pro feature with its own tab.
How is that different from MIDI Learn?
MIDI Learn binds one control to one parameter and that is the whole relationship. It cannot smooth a jittery stick, hold a value after you let go, fire only on a double tap, snap a sweep to a musical scale, or modulate a parameter while your hands are elsewhere. The node graph is for the behaviour between the input and the output, which Learn has no way to express.
How many nodes are there?
Seventy-nine built in, grouped into six categories: 21 shape nodes, 15 modulators, 12 gates, 11 logic nodes, 10 spatial nodes for XY work, and 9 music nodes. The registry also loads nodes from a user folder, so the set is extendable.
Do I need to understand modular synthesis to use it?
No. The useful starting patches are two or three nodes long — a dead zone to kill stick drift, a smooth node to stop a jittery CC, a scale quantize so a stick sweep only lands on notes in key. The modular depth is there if you want it, not a prerequisite.
Does the graph run in real time without adding latency?
It is evaluated once per bridge tick, walking the graph in topological order, at the same poll rate as the rest of the bridge. Cycles are refused at evaluation time rather than allowed to spin.
Where is the graph saved?
In the mapping itself, so it travels with the preset. Load that preset in another session and the patch comes with it rather than being stored loosely in application state.