Conditioning Merge (Multi)

Same job as Conditioning Merge, more inputs. This one blends up to five conditionings into one, which matters the moment you're feeding several reference-image conditionings into a Krea 2 or Ideogram 4 workflow and don't want to daisy-chain a stack of two-input merges. One node, up to five sources, one output. Cleaner graph, easier to reason about.

The reason it exists in this pack: the encode and edit nodes each produce a conditioning, and a multi-character or multi-reference edit means you've got several of them to combine. Chaining five separate Conditioning Merge nodes works but turns your canvas into spaghetti and makes the actual blend ratios hard to follow. This collapses that into one node.

Inputs and outputs

Required:

• conditioning_1 - the first, always-present input.
• match_percent - default 0.5, range 0 to 1. The blend ratio governing how the inputs are weighted together. 0.5 is an even mix; nudge it to shift the balance.

Optional:

• conditioning_2 through conditioning_5 - plug in as many as you need, up to four more. Leave the rest empty and it merges however many you actually connected.

Output is a single conditioning, ready for your sampler's positive input or another node downstream.

The practical upshot: connect two to five conditionings, set one ratio, get one blended result. You don't wire per-input weights - there's a single match_percent, so this is a "mix these evenly-ish" tool, not a per-source mixing desk. If you need one source to dominate a fixed reference, look at Conditioning Merge (Anchor) instead.

Installing it

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

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

Restart ComfyUI. No pip deps and no model files - it's pure conditioning math and model-agnostic, though it ships in a Krea 2 / Ideogram 4 pack.

Common issues

Keep every input in the same model's conditioning space. Mixing conditionings from different base models is the reliable way to get mush or an error - this node assumes they're compatible and won't warn you if they aren't.

Watch what you actually connected. Because conditioning_2–conditioning_5 are optional, it's easy to think you merged five things when a wire quietly popped off and you merged four. If a result looks off, count your connected inputs before touching match_percent.

And remember it's one ratio for the whole blend. If your real goal is "reference A locked, B and C blended around it," this node can't express that cleanly - that's exactly what the anchor variant is for. For a straight "combine these few conditionings," though, this is the tidy choice and beats a chain of two-input merges every time.