Sqrt (Audio Op)

Straight to the point: OpSqrt does not compute a square root. Look at the source and you'll find:

def sqrt(y, sr):
    return y**-2

That's inverse square, not square root. It shipped with the "Sqrt" display name anyway, probably because the author (dmarx, porting his video-killed-the-radio-star notebook) jotted the math down fast and never came back. So before you reach for this thinking it'll soften a curve the way smoosh does, know what you're actually getting.

What does y**-2 do to a signal? It's the inverse of pow2, and it's aggressive. Values below 1 (which is nearly everything in an audio signal) blow up when raised to a negative power: 0.5**-2 = 4, 0.1**-2 = 100. Near-silence becomes enormous. Since y in a SIGNAL is typically a normalized waveform in roughly [-1, 1], this produces huge, inverted-looking numbers - unless the signal is strictly positive (e.g., an RMS envelope), in which case it's a violent contrast flattener: it crushes the peaks toward 1 and yanks the quiet parts up to massive values.

Honest advice: this is the node in the pack most likely to make you think your graph is broken when it isn't. If you actually wanted a square root - the gentle compression smoosh does - grab OpSmoosh instead, which does normalize-then-y**0.5-then-normalize and is clearly the thing "Sqrt" was probably meant to be. If you genuinely want inverse-square contrast shaping, this works, but you'll usually want Normalize right after so the output stays in a usable range.

Inputs and outputs

• signal (SIGNAL) - from ARReadAudio or upstream of another operator.
• SIGNAL - y replaced by y**-2, same length as the input. Feed it to ARDrawSignal before you trust it, because the numbers will look weird even when it's doing what it says.

Installing it

Manager (search "AudioReactive") or:

cd ComfyUI/custom_nodes
git clone https://github.com/dmarx/ComfyUI-AudioReactive

Restart and wait out the first-load auto-install of scipy, scikit-learn, librosa, loguru (librosa pulls numba). If the pack won't load with ModuleNotFoundError: No module named 'keyframed', pip install keyframed.

Common issues

• Output numbers are enormous - that's y**-2 being y**-2. Follow with Normalize or pick Smoosh.
• You expected a square root - this isn't one. That's not you misunderstanding the node; the node is mislabeled.
• Curve blows up to infinity near zero - raising near-zero values to a negative power diverges; the floating-point result can be huge (or inf). If the pack's normalize were applied you'd be fine, but this op doesn't normalize.
• General early-port caveat - this op works (it computes what the source says), but the pack is a fresh single-commit port with a one-line README and several parameterized operators left as unwired stubs. Read the source before trusting display names around here.