Conditioning Add ConDelta

The README's ELI5 walks you through the file-based version of this pack: save a ConDelta, drop it in models/ConDelta, pick it from a dropdown with ApplyConDelta. That's a fine workflow once you know what delta you want. It's a lot of friction while you're still figuring that out. When the pack's own launch thread was asked whether you have to save a delta every time, the answer from the community was blunt: "No need to save them actually, you can generate and apply on the fly." ConditioningAddConDelta is that on-the-fly node - it takes a delta straight from whatever built it (a subtract, an average, GetConDeltaFromPrompt) and adds it directly to a base conditioning, no file system involved.

How it works

Same underlying operation as ApplyConDelta - add a scaled delta to a base conditioning - but where ApplyConDelta loads its delta from a saved file, this node takes it as a live CONDITIONING input from anywhere upstream in your graph. That makes it the right node once you're iterating: build a delta with a subtract or with GetConDeltaFromPrompt, wire it straight in here, generate, tweak the upstream prompt, generate again - no save step slowing down the loop.

The inputs and outputs that matter

• conditioning_base - the conditioning you're modifying.
• conditioning_delta - the delta itself, from anywhere in your graph. Native ConditioningSubtract plus this node is the manual equivalent of what SaveConditioningDelta + ApplyConDelta gives you after a save/reload round trip.
• conditioning_delta_strength - default 1, range -100 to 100. Positive applies the delta's concept, negative pushes away from it.

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 setup needed for this node - that's the whole advantage over the file-based nodes.

Common issues & troubleshooting

This is the node for experimentation; save for reuse. Once you've landed on a delta that actually works, route it through SaveConditioningDelta so it shows up in ApplyConDelta's dropdown for future workflows instead of rebuilding the subtract chain every time. Use this node while you're still exploring, switch to the file-based version once you've found something worth keeping.

Strength behaves the same as everywhere else in the pack - too much burns the image. The ±100 range is generous; the useful range is usually a small fraction of that. If you want the strength to auto-normalize against the base conditioning's own magnitude instead of guessing, use ConditioningAddConDeltaAutoScale instead - same node, plus the ratio-based scaling from ApplyConDeltaAutoScale.

Keep the delta and the base conditioning on the same model family. A delta computed from one architecture's CLIP or T5 output isn't a meaningful thing to add to a different architecture's conditioning - same caveat the README gives for saved deltas applies here even though nothing's been saved.