Humanizer by humanize: Pink Noise Timing for Stiff MIDI
Most timing randomisers roll a fresh dice for every note. Nudge a hi-hat two milliseconds late, the next one three early, the one after that back on the grid. It stops the sequence sounding like a machine, but it does not sound like a person either, because people do not drift at random. Their errors are correlated: a slightly late note tends to be followed by another slightly late note, and the whole phrase wanders.
Humanizer by humanize is built on exactly that distinction. Each incoming MIDI note is delayed by a random amount, but the random values come from pink noise — 1/f noise — rather than from white noise. The result is a drift with memory in it. The device credits the research it is based on, Holger Hennig et al., "The Nature and Perception of Fluctuations in Human Musical Rhythms" (PLoS ONE, 2011), and the Max implementation is attributed to James Holden.
What "1/f" actually changes in the sound
Think of two drummers. One has a nervous twitch: every hit lands in a slightly different place with no relationship to the previous one. The other pushes and pulls across bars — a little ahead through the fill, settling back behind the beat for the next eight. The first is white noise. The second is closer to 1/f, and it is what Humanizer's pink~-driven delay is reaching for.
Practically, that means the device is more interesting over long passages than over two bars. Loop four bars of a part through it and you will hear small deviations. Let sixteen or thirty-two bars play and you start hearing the slower wander, which is the part that reads as a performer rather than as jitter.
Where it sits in the chain, and the one setup detail to get right
It is a MIDI effect, so it goes in a MIDI track's device chain ahead of the instrument. Anything you put after it — arpeggiators, chord devices, scale correction — will receive the already-displaced notes, which matters if you are chaining several MIDI devices. Humanize last, just before the instrument, if you want the deviations applied to the final note stream rather than to a seed that something else then multiplies.
The important detail: the device delays notes. It cannot pull a note earlier than it arrives, because at the moment the note arrives it has no way of knowing the future. If you want the part to sound like it sits both ahead of and behind the grid rather than consistently behind it, set a negative Track Delay on that MIDI track to pre-compensate, so the average landing point is back on the beat and the drift goes both ways around it. Audition this against the dry part rather than guessing the number.
A second consequence of live processing: the displacement is generated as the notes pass through, so each pass is a new performance. If you need one specific take to be permanent, resample or record the MIDI out to another track.
Three sessions where it earns its slot
Programmed drums that refuse to breathe. You have a kit built from one-shots, everything quantised to 100%. Groove Pool templates will shift notes, but a groove is a fixed pattern that repeats identically every loop. Humanizer gives you non-repeating variation instead. Put it on the kit track and run a long section; if the whole kit moving together is too much, split hats onto their own MIDI track with their own instance so the hats drift independently from the kick and snare.
Quantised keys and arpeggios. A Rhodes or plucked synth part played in and then hard-quantised loses the micro-timing that made the original take feel like hands. Humanizer on that track restores timing variation after the fact, which is useful when you quantised for pitch accuracy reasons and regretted the side effect. Chords are where it shows up most clearly — the notes of a chord no longer land simultaneously, which is close to what fingers actually do.
Layered sequences that are too tight against each other. Two synths doubling the same line sound like one synth with a chorus on it when their note-ons are sample-aligned. An instance on each track separates them. Be aware that two independent instances drift independently, which is a different thing from two musicians listening to each other — and that limitation is the reason the author wrote a follow-up.
Read the supersession note before you download
The device page opens with a notice in capitals from the author: Humanizer has been superseded by the Group Humanizer, which handles multiple channels and keeps their variations correlated with each other, and the author recommends downloading that one instead. That is unusually direct, and worth taking at face value. If your use case is a single track, this device does the job. If it is a drum kit split across tracks, or an ensemble where the parts should drift together the way players following each other would, the newer device is the one designed for it.
The other practical caveat is age. Humanizer is at version 1.1, updated in March 2015, and was built against Live 9.0.6 and Max 6.1.4. Max for Live devices from that era often still load, but if you are on a current Live version, treat compatibility as something to check rather than assume. The licence is Attribution-NonCommercial-ShareAlike, and the page describes the implementation as free for private use — a point to weigh if the track is going out commercially.
The numbers, and what they do and don't say
Just under 20,000 downloads and a 5/5 rating, accumulated since January 2014. Spread over a decade that is steady, unglamorous demand — the profile of a utility people find when they go looking for a specific solution, not something that trended. Five comments is a small discussion thread, so read the 5/5 as "nobody who rated it was disappointed" rather than as a large consensus. Combined with the author's own note pointing elsewhere, the honest summary is: a well-regarded, narrowly-scoped tool whose successor is probably what you want.
Its neighbours in the MIDI effect catalogue
Devices sitting near Humanizer on popularity in the MIDI effect list solve different problems, which is a useful way to see what Humanizer is not.
- Tombola Sequencer by joelpowell generates the notes instead of adjusting them — up to 64 notes bouncing in a rotating shape with Speed, Mass, Gravity and Shape, all automatable. Reach for it when you want unpredictable material, not unpredictable timing. There is a fuller look at it in the Tombola deep dive.
- Stepper by dric is a step sequencer and LFO hybrid, up to 16 steps with per-step waveforms, aimed at modulating any parameter in Live. Deterministic where Humanizer is stochastic.
- CC map by veedjee is a bidirectional CC controller for driving hardware, with MIDI learn — a routing tool rather than a feel tool. See the CC map deep dive if hardware control is the actual problem.
- MotionMod by zealtv turns camera motion into modulation. It is the other device on this list that injects human physicality into a session, but through a webcam rather than through statistics.
Who should skip it
Skip Humanizer if you need correlated timing across several tracks — that is what the Group Humanizer exists for. Skip it if you want velocity variation: the behaviour described is timing displacement, and velocity humanisation is a separate job. Skip it if you are unwilling to troubleshoot a device built for Live 9 and Max 6. And skip it if the part in question is a tight electronic groove where grid-locked timing is the point; drift is not an improvement everywhere.
Keep it in mind if you work with programmed acoustic kits, quantised keyboard parts, or long evolving sections where the slow wander has room to be heard — and where one track, not six, needs to stop sounding like a sequencer. For wider context on what else is available for reshaping note data, the MIDI transformation roundup covers the neighbourhood.