Conditioning Average Extended

ComfyUI ships a native "Conditioning (Average)" node, and it's fine - it blends two conditionings with a weight clamped to 0–1. That clamp is the whole reason this node exists. Extended Conditioning Average does the same weighted blend but removes the leash: conditioning_to_strength runs from -100 to 100, so instead of only ever landing between your two conditionings, you can push straight through one of them and out the other side.

That distinction matters more than it sounds. A 0–1 blend can only interpolate - find a point on the line segment between "a photo of a city" and "an art deco city." An extrapolated blend can go past the endpoint, exaggerating the direction from one toward the other rather than just diluting it. It's the same idea the rest of this pack's ConDelta math leans on: treating conditioning as a vector space you can push around in, not just a fixed point you pick from a menu.

How it works

You give it two conditionings - conditioning_to and conditioning_from - and a strength. At conditioning_to_strength = 1 you get conditioning_to unchanged; at 0 you'd get closer to conditioning_from; at 0.5 you land in between. Values outside that range extrapolate past conditioning_to, amplifying whatever direction it represents relative to conditioning_from.

The inputs and outputs that matter

• conditioning_to and conditioning_from - your two encoded prompts, no restrictions beyond both needing to come from the same model family.
• conditioning_to_strength - default 1, range -100 to 100, step 0.01. This is the only real knob on the node, and it's doing double duty as both a blend weight (inside 0–1) and an extrapolation factor (outside it).

Output is a single CONDITIONING, ready to feed into your sampler like any other conditioning.

How to install it

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

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

then restart. No model downloads, no extra dependencies - this node is pure arithmetic on conditioning tensors you already have.

Common issues & troubleshooting

Stay close to the sane range at first. Interpolation between 0 and 1 is well-behaved by construction - you're always somewhere between two real prompts. Extrapolation isn't: push past 2 or 3 and you're amplifying a direction with nothing constraining the result, and you'll see the same kind of burn and structural breakdown that comes from cranking CFG too high. Creep outward from 1 in small steps rather than jumping to a big number.

If you only need a plain 0–1 blend, use the native node. ComfyUI's built-in Conditioning (Average) does exactly that job and is one fewer custom node to maintain. Reach for this Extended version specifically when you want to push past an endpoint rather than just sit between two prompts - for example, exaggerating a style direction one prompt has relative to a plainer baseline, which is conceptually the same move as a ConDelta, just phrased as an average instead of a subtract-then-add.

Order matters. Which conditioning is _to and which is _from decides which direction you're extrapolating in when you go outside 0–1 - swap them if the result is pushing the wrong way.