Conditioning Add ConDelta AutoScale

This node sits at the intersection of two ideas the rest of the pack already gives you separately: ConditioningAddConDelta's on-the-fly, no-file-required workflow, and ApplyConDeltaAutoScale's magnitude normalization. If you're iterating on a delta live - built from a subtract, an average, or GetConDeltaFromPrompt - and you're tired of re-tuning strength every time the delta's own scale changes, this is the node that saves you the trouble.

How it works

It adds conditioning_delta to conditioning_base, same as the plain add node, but first normalizes the delta's magnitude relative to the base conditioning using whichever statistic ratio_type selects - mean, max, or median - before applying conditioning_delta_strength. That means a strength of 1 means roughly the same thing regardless of how large or small the raw delta happens to be, which matters a lot when the delta is coming fresh out of an average or a subtract rather than a hand-tuned saved file. There's also an optional clamp: turn it on and the scaled delta's values get capped at clamp_value, a backstop against a delta that turns out much larger than expected once normalized.

The inputs and outputs that matter

• conditioning_base - the conditioning you're modifying.
• conditioning_delta - the live delta, from anywhere upstream.
• ratio_type - default median, with mean and max available. Median is the safer starting point; it's less skewed by outlier values in the delta than mean or max.
• conditioning_delta_strength - default 1, range -100 to 100, applied after scaling.
• clamp - default off; the tooltip is explicit: "If true, will clamp the ConDeltas's values."
• clamp_value - default 3, only matters when clamp is on.

Output is a single CONDITIONING.

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 folder needed here either - everything comes from live graph connections.

Common issues & troubleshooting

Default to this over the plain add node once you're comparing multiple deltas. If your workflow swaps between deltas built from different sources - one from GetConDeltaFromPrompt, one from a manual subtract, one from ConditioningAverageMultiple - the auto-scaling here means you don't have to re-guess a sane strength for each one individually.

If burn shows up at a strength that felt safe before, that's the normalization changing the effective scale - turn clamp on rather than hunting for a new strength value by trial and error. It's a cheaper fix than re-tuning.

Try both mean and max if median doesn't feel right for a particular delta. They can genuinely disagree on a delta with an uneven value distribution, and there's no cost to switching between them to see which one gives you the behavior you expected.