The problem
The photographic negative of a mix's dynamics. Per band, whatever sat above that band's own running average is pushed down by as much as it was up, and whatever sat below is lifted by as much as it was down. Time order is untouched — the loop still plays forwards and still sounds like itself, turned inside out.
Dynamics tools move in one direction. A compressor narrows the distance between loud and quiet, an expander widens it, and both leave the same material in front. Reflecting that distance is a different operation from narrowing or widening it.
The two nearest tools are not it either. A transient shaper subtracts the attack and leaves the sustain where it was, which is the same performance with less front end rather than its inverse. A downward expander pushes quiet material further down, which is the opposite of the interesting half. What is left is taking the space between the hits and making it the subject.
How it works
01
The pivot is an average, not a threshold
Each band reflects around its own slow running average. A threshold has to be found again for every source and every gain staging; an average means the knob does the same musical thing on a −30 dBFS loop and on a mastered file, with nothing to set up first.
02
Sixteen bands, reflected separately
A 43 ms analysis window fixed in time rather than samples, so the sound does not change with the session rate, split into 16 bands from 30 Hz to 18 kHz. They peak at different moments, so when the kick is pushed down the hats, the tail and the room come up into the hole it leaves — which is why it survives on a full mix instead of pumping the whole thing.
03
The midpoint is a destination
Amount runs dry at 0, fully reflected at 100, and at 50 every band is pinned to its own average — everything equally loud. That middle setting is a levelling machine with no threshold and no attack time, and it is a place to stop rather than somewhere passed through.
Numbers
Every figure comes from a test
Specification
- Analysis
- 43 ms window, 75% overlap, fixed in time
- Bands
- 16, from 30 Hz to 18 kHz
- Latency
- 2048 samples at 44.1 and 48 kHz, 4096 at 96
- Presets
- 8
- Formats
- VST3, CLAP, LV2
- Systems
- macOS (Apple Silicon and Intel), Windows 64-bit
Measured
- Per-band inversion, Depth 36 dB
- −0.982 correlation
- Per-band inversion, Depth 18 dB
- −0.945 correlation
- Broadband loudness against dry
- +0.026 — decorrelated, not inverted
- Reconstruction at Amount 0
- −136.5 dB at 44.1 kHz
- Band spread at Amount 50
- 6.1 dB, against 12.7 dry
- Latency, reported against measured
- exact at 44.1, 48 and 96 kHz
Presets
Inside Out, Ghost Room, Flatten, Reverse Gravity, Dust Lift, Undertow (Parallel), Barely There, Total Negative
Free
Checking what is available…
The others
They are built the same way
Negative is free and unlicensed: it asks for no address, creates no account, and reports nothing to any server. The comparisons above are to categories of processing — compressors, expanders, transient shapers — and not to any particular product. Nothing here is measured from, matched to or named after anything on sale.