Conditioning Merge (Anchor)

The merge node with an opinion. Where Conditioning Merge (Multi) treats all its inputs as peers, Conditioning Merge (Anchor) designates one conditioning as the anchor - the fixed reference everything else blends around rather than into. If you've got a character or subject you want held steady while you mix in style, pose, or secondary references, this is the node that keeps the anchor from getting diluted.

That's a real need in this pack's world. Multi-reference Krea 2 / Ideogram 4 edits tend to have one thing you care about most - a specific face or subject - and several supporting references you want to influence the result without overwriting the main one. A flat merge averages them all and the important one drifts. An anchored merge treats it as the baseline.

Inputs and outputs

Required:

• anchor - the conditioning you want to hold as the reference frame. This is the thing that stays put.
• conditioning_1 - the first blend-in conditioning.
• match_percent - default 0.5, range 0 to 1. How much the other conditionings are allowed to pull the result away from the anchor. Low keeps things close to the anchor; higher lets the extras assert themselves.

Optional:

• conditioning_2 through conditioning_5 - more conditionings to blend in around the anchor, up to four more.

Output is a single conditioning for your sampler's positive input or further downstream nodes.

The mental model: anchor = the subject you're protecting, the numbered inputs = everything modifying it, match_percent = how loud those modifiers get. Keep match_percent low when the anchor's identity matters and you're only seasoning it; raise it when you want a genuine blend and the anchor is just the starting point.

Installing it

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

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

Restart ComfyUI. No pip dependencies, no model files - pure conditioning math, model-agnostic, though it belongs to a Krea 2 / Ideogram 4 pack.

Common issues

Same-space rule as the other merges: the anchor and every blended input must come from the same base model's conditioning. Mix a Krea conditioning with an SDXL one and the result is meaningless.

The most common mistake is expecting the anchor to be immovable. It isn't a lock - it's the reference the blend is measured against, and a high match_percent will still drag it. If your anchored subject keeps drifting, pull match_percent down before you blame the node.

This is the most niche merge in the pack (its page barely registers any search traffic), and honestly you only need it when a flat multi-merge is washing out your main subject. For most blends, Conditioning Merge or Conditioning Merge (Multi) is simpler and does the job - reach for the anchor variant specifically when preserving one reference against several others is the whole point.