Average Multiple Conditionings or ConDeltas

ComfyUI's native averaging tools work two conditionings at a time. Chain enough of them together and you can average any number you want, but the graph gets ugly fast - nine average nodes to blend ten prompts. ConditioningAverageMultiple does it in one: plug in anywhere from one to ten conditionings and get a single averaged result back.

How it works

conditioning0 is required; conditioning1 through conditioning9 are all optional. Wire in as many as you need - the node averages together whatever's actually connected. There's no per-slot weight exposed on this node, which is worth knowing going in: everything you plug in contributes equally to the result. If you need an uneven blend, where one prompt should count more than the others, pre-scale it with ConditioningScale before feeding it into one of the slots here.

Why you'd reach for it

The obvious use is exactly what it sounds like - blending several style or subject prompts into one averaged conditioning. It's also the building block this pack's own GetConDeltaFromPrompt almost certainly leans on internally, since building a "baseline" for its misc or anime categories means averaging a pool of generic prompts together before subtracting. If you ever want to build your own baseline pool by hand - say, for a custom delta workflow, or just to establish "what does an average prompt in this style look like" as a reference point - this is the node for it.

The inputs and outputs that matter

• conditioning0 - required, the one slot you must fill.
• conditioning1 through conditioning9 - optional, fill in as many as you're blending.

Output is a single CONDITIONING, the average of every connected slot.

How to install it

Via ComfyUI Manager: search "ComfyUI-ConDelta", install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/envy-ai/ComfyUI-ConDelta

then restart. No model files needed for this node.

Common issues & troubleshooting

No weighting means dominant prompts can get diluted. If one of your ten conditionings represents the concept you actually care about and the other nine are supporting texture, equal-weight averaging will pull the result toward the group rather than your intended focus. Pre-scale the important one up with ConditioningScale before it reaches this node, or drop the count down to fewer, more deliberately chosen inputs.

Averaging more prompts tends to produce a "muddier" result, not a sharper one. This is the nature of averaging in vector space rather than a bug - each additional prompt pulls the result toward the group centroid, which is useful for building a generic baseline but not for combining distinct strong concepts. If the output feels washed out, you're probably averaging too many things that don't actually share a direction.

If the blend comes out with an unexpectedly large or small magnitude, run it through ClampConDelta or ApplyConDeltaAutoScale's normalization before using it further down the chain - averaged conditionings don't come with any guarantee about their resulting scale.