Multiply Conditioning Strength

Conditioning in ComfyUI is a set of text-embedding tensors that steer the sampler. Normally, if you want a prompt to have more influence on the image, you fiddle with the words or bump CFG. Multiply Conditioning Strength does something more surgical: it takes a conditioning and literally scales it by a number, so a whole prompt branch gets louder or quieter without you touching a single word.

That "scale a conditioning" operation is a real thing people reach for - it's the raw-material version of prompt weighting, applied at the embedding level instead of with syntax tricks. If you've ever wanted "this part of the prompt should matter 1.5× more," this is the honest way to do it.

How it works

Two inputs:

• conditioning - the conditioning to scale, straight from a CLIP Text Encode.
• operation - a dropdown that picks multiply, divide, or none, each with a value field. Multiply with value = 1.5 scales the conditioning up by 1.5; divide does the inverse; none passes it through untouched.

The output is the scaled conditioning, which wires back into whatever consumed the original - a KSampler's positive or negative input, a conditioning combine node, a ControlNet's conditioning slot. Because it's a simple tensor scaling, it composes cleanly: you can scale a combined conditioning, or scale branches before combining them.

What the value actually does

This is where you should keep expectations honest. Scaling conditioning up intensifies the influence of those embeddings on generation - think of it as a blunt force version of prompt weight, with real effect on how strongly a subject or style asserts itself. But it's not a magic "make it better" knob: too high and you can overdrive the branch (colors oversaturate, composition fights itself); too low and the branch fades toward ignored. Start around 1.0–1.2 for a subtle lift, and treat anything past 1.5 as territory that needs visual testing per checkpoint.

Where it earns its keep

The two classic uses: negative-prompt tuning (scaling the negative conditioning up or down is a more controllable lever than cranking CFG, and it's the thing people actually mean when they say "stronger negative") and branch balancing - when you've combined a subject prompt and a style prompt into one conditioning, scaling each side before combining gives you per-theme control instead of one global CFG. It's a utility, but it's the kind of utility that unblocks workflows you thought needed a different sampler.

Installing

It ships in Sage Utils:

cd ComfyUI/custom_nodes
git clone https://github.com/arcum42/ComfyUI_SageUtils.git
cd ComfyUI_SageUtils && pip install -r requirements.txt

Restart ComfyUI, or install "Sage Utils" via ComfyUI Manager. No models, no dependencies beyond the pack.

Where people get confused

The divide mode's value field has a small floor (it won't accept 0, because dividing by zero is how you summon the void) - so "make it silent" is done with a very small multiply, not a zero divide. And a reminder about modern models: a lot of the flow-matching checkpoints are guidance-distilled and want CFG at 1, where the negative branch has little to attach to. Conditioning scaling still works on the positive side there, but if you're used to negative-conditioning tricks from the SD 1.5 era, expect them to behave differently on those models. Test per model, as with everything.