Conditioning Merge

The plain-vanilla blender of the pack. Conditioning Merge takes two conditionings and mixes them into one, with a single ratio controlling how much of each you get. If you've ever wanted "60% this prompt, 40% that one" - or to blend a rebalanced reference conditioning back toward your plain text prompt so it doesn't run away - this is the node.

It lives in the same family as ComfyUI's built-in conditioning combine/average nodes, and if you already know those, you know the shape of this. The reason it ships inside this pack is that the rebalance and encode nodes throw off conditionings you often want to temper - mix a heavily-rebalanced reference conditioning with a clean text one and you keep the reference influence without letting it eat the whole image.

Inputs and outputs

Three inputs, all required:

• conditioning_1 - the first conditioning.
• conditioning_2 - the second.
• match_percent - default 0.5, range 0 to 1. The blend ratio. At 0.5 the two are weighted evenly; slide it toward 0 to lean on conditioning_1, toward 1 to lean on conditioning_2. That's the whole control surface.

Output is a single conditioning that wires wherever a conditioning goes - usually your sampler's positive input, or into another merge/rebalance node if you're stacking.

There's nothing hidden here. It's a two-input crossfader. The value is in what you feed it, not in the node itself.

Installing it

ComfyUI Manager: search ComfyUI-Conditioning-Rebalance, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/nova452/ComfyUI-ConditioningKrea2Rebalance.git

Restart ComfyUI. No pip dependencies, no model files - this one's pure tensor arithmetic and doesn't care which base model you're on, though it lives in a pack built around Krea 2 and Ideogram 4.

Common issues

The usual failure isn't the node, it's mismatched conditionings. Merging conditionings from two different models - an SDXL one and a Krea 2 one, say - is asking for trouble; keep both inputs in the same model's space. Within one model, blending is fine and predictable.

If a blend looks like it's ignoring one input, check you didn't leave match_percent pinned at 0 or 1 - at the extremes you're effectively passing one conditioning straight through and wondering where the other went.

For blending more than two, reach for Conditioning Merge (Multi) (up to five inputs) or Conditioning Merge (Anchor) (five plus a fixed anchor) instead of chaining a dozen of these - the multi-input versions are cleaner and easier to reason about. Otherwise there's not much to go wrong: it's the simplest, most reusable node in the pack, and the one you'll drop into graphs long after the reference-editing experiments are done.