Conditioning Strength Scaler (Arctenox's Essentials)

Your negative prompt is doing too much - it's scrubbing away detail you actually wanted - and you'd rather not retype a paragraph of tags to find out. This node is the global volume knob for a conditioning: it multiplies the CONDITIONING tensor by a scalar before the sampler ever sees it. Soften a negative to 0.8, push a positive to 1.3, or zero a branch out entirely to see what it was contributing.

How it works

A CONDITIONING in ComfyUI isn't one number; it's a list of (tensor, metadata) pairs, where the metadata carries things like the timestep range the conditioning applies to. This node loops over those pairs and scales each tensor by strength, keeping the metadata intact.

The useful bit of the implementation: at exactly 1.0 it returns the original object untouched - no copy, no math, genuinely zero cost. Everywhere else you pay for a scaled tensor, which is cheap.

The strength input runs from -4.0 to 4.0 in 0.05 steps, default 1.0. Above 1.0 amplifies, below 1.0 dampens, 0.0 zeroes the conditioning out (no guidance from it), and negative values invert the direction of guidance.

What you'll actually do with it

Three moves, in order of how often people reach for them:

• Dampen the negative. Drop a scaler on your negative conditioning at 0.7–0.9 when it's fighting the positive. It's the gentlest way to back a prompt off.
• A/B test a branch. Duplicate a conditioning, scale one copy, and see how much of the result it owns. Way faster than rewriting and re-encoding.
• Boost a phase. The pack's own Prompt Phase Splitter produces weighted conditionings you can tune downstream with this node instead of rebuilding them.

The traps

Amplifying hard is the easy mistake. Conditioning is a gradient direction, and pushing it to 2.0+ behaves a lot like raising CFG - saturated, "burnt" colors, blown-out contrast, the exact overshoot you're probably already fighting. If you're used to CFG values, think of this as a per-conditioning CFG dial, not a magic sharpening control.

And negative strengths? They invert the guidance direction, which in practice usually produces garbage. It's an experiment knob, not a setting. Leave it in positive territory unless you're deliberately probing what a conditioning encodes.

Install

Standard pack install - ComfyUI Manager, search Arctenoxs-Essentials_ComfyUI, or:

cd ComfyUI/custom_nodes/
git clone https://github.com/Arctenox/Arctenoxs-Essentials_ComfyUI

then restart. No model downloads, nothing heavy. One line of context: the README flags the pack as deprecated while the author folds it into a new pack, so don't expect a flood of updates - but this node is a self-contained ten lines of logic and will keep working on current ComfyUI for the foreseeable future.