⚙️ CR Conditioning Mixer

ComfyUI already ships three separate nodes for merging conditioning - Conditioning (Combine), Conditioning (Average), and Conditioning (Concat) - and if you've ever wanted to A/B them you know the annoyance: you have to physically swap nodes to change the method. CR Conditioning Mixer folds all three into one, with a dropdown. Pick the method, done.

That's the whole pitch, and it's a good one for anyone building blended prompts, doing prompt-morph experiments, or just keeping a graph tidy. You're not getting new math here - you're getting a switch over the math ComfyUI already has.

How it works

Conditioning in ComfyUI is the encoded output of a text prompt (or an IP-Adapter, or a ControlNet apply). "Mixing" two of them means deciding how the sampler sees both at once, and the three methods do genuinely different things:

• Combine keeps both conditionings as separate guidance signals - the sampler is steered by each in parallel. Good for jamming two distinct ideas together without one erasing the other.
• Average interpolates the two tensors into one, weighted by strength. This is your prompt-blend / prompt-morph lever.
• Concatenate literally joins the token sequences end to end, as if you'd written one longer prompt.

The inputs that matter

• conditioning_1 / conditioning_2 (CONDITIONING) - the two encoded prompts to merge.
• mix_method - the dropdown: Combine, Average, or Concatenate.
• average_strength (FLOAT 0–1, default 0.5) - the blend weight. It only does anything in Average mode; at 0.5 you get an even mix, push it toward 1 to favor conditioning_1. In Combine and Concatenate it's ignored.

Output is a single CONDITIONING that goes straight into your KSampler's positive (or negative) input, plus a show_help string you can ignore.

Installing it

This is a Core node in Comfyroll Studio (Suzie1 / RockOfFire), a long-running utility pack most people already have. Via ComfyUI Manager: search "Comfyroll Studio," install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/Suzie1/ComfyUI_Comfyroll_CustomNodes.git

then restart. No models or extra dependencies.

Common issues

The most common confusion is average_strength "not doing anything" - that's because you're in Combine or Concatenate mode, where it's supposed to be inert. Switch to Average.

Beyond that, mixing conditioning is a blunt instrument. If you want two prompts in two regions of the image, this isn't the tool - you want proper regional prompting with masks. Concatenate can also blow past CLIP's token window if both prompts are long, which quietly truncates. And remember both inputs must be encoded against the same model's CLIP; you can't average an SDXL conditioning with an SD 1.5 one. If the pack fails to load on startup with a traceback, reinstall it cleanly through Manager rather than debugging the node itself.