ACE-Step 1.5 Keystone Config (Experimental) ⚡🅡🅞🅣🅘

This node doesn't do anything on its own - it's a config object, not an effect. What it holds is genuinely interesting, though: six specific channels in ACE-Step 1.5's latent space that this pack's author identified, through what the tooltips describe as "Phase 1" experiments, as having outsized individual impact when perturbed, each ranked by an actual measured impact score rather than a guess. It's marked "(Experimental)" in its own display name, and the tooltips are unusually candid about that - several channels are flagged as having effects that are "content-dependent and under active investigation," not settled behavior.

How it works

There's no audio, model, or latent input here at all - just six gain sliders, one per keystone channel, each defaulting to 1.0 (neutral): ch19 (impact score 107, the highest of the six, sensitivity 1.2), ch29 (impact 99, second highest, and specifically noted for producing the largest single-channel spectral flatness shift, sensitivity 1.2), ch56 (impact 91, inverted, sensitivity -1.0), ch13 (impact 74, inverted, sensitivity -0.7), ch23 (impact 27, described as showing effects "at all gain levels," sensitivity 1.0), and ch14 (impact 32, the subtlest of the six, "primarily noticeable at extreme gains," sensitivity 1.5).

The output is a KEYSTONE_CONFIG object - it does nothing by itself. You connect it to the keystone_config input on either this pack's Generation Steering node (to steer what ACE-Step generates, during sampling) or its Latent Channel EQ node (to scale the model's output after the fact), where it multiplies on top of whatever those nodes' own broader band gains are already doing. Think of the six band gains on those other two nodes as coarse tone controls, and this node's six individual channels as the fine, surgical adjustment layered on top.

Inputs and outputs

• The six channel sliders (ch19, ch29, ch56, ch13, ch23, ch14) - each 0 to 2, neutral at 1.0. That's the entire input surface.
• Output - a single KEYSTONE_CONFIG, which does nothing until you wire it into a node that consumes it.

There's genuinely nothing else to configure - no model input, no strength master, no windowing. All of that timing and overall-strength control lives on whichever downstream node (Generation Steering or Latent Channel EQ) you connect this to.

Installing it

Install through ComfyUI Manager (search RyanOnTheInside) or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/ryanontheinside/ComfyUI_RyanOnTheInside

Then pip install -r requirements.txt and restart ComfyUI. This node has no dependency of its own beyond the pack as a whole, but it's useless in isolation - you need an ACE-Step 1.5 pipeline and one of the two downstream nodes it feeds already set up.

Common issues

The most common misunderstanding is expecting this node to do something visible on its own - it won't; if you build a graph with only Keystone Config and nothing consuming its output, nothing will change about your generation. Wire it into Generation Steering or Latent Channel EQ's keystone_config input before expecting any effect.

Given the two inverted channels (ch56, ch13), the same warning applies here as on Generation Steering's inverted bands - moving those sliders up from 1.0 pushes the result in the opposite direction from the other four channels, which trips people up if they're adjusting all six sliders by the same intuition. And because the author's own notes describe several of these effects as content-dependent and still under investigation, treat this as a genuinely experimental tool for exploration rather than a documented, reliable mixing control - small moves, listen, adjust, rather than dialing in a target value from a formula.