CLIP Encoded Prompt Bus

The Plain Prompt Bus passes prompt strings. This one is what you use when the prompts have already been run through CLIPTextEncode - because by then you're holding three values that all want to end up at the KSampler: the model, the positive conditioning, and the negative conditioning. The CLIP Encoded Prompt Bus bundles exactly that trio, and its README entry is unambiguous about it: "CLIP Text Encoded positive and negative prompt and associated model."

So the typical shape is: CLIPTextEncode (positive) and CLIPTextEncode (negative) in parallel, both taking the clip from your loader, and the model coming straight from the checkpoint - then all three fed into this bus, which hands them back out in one spot right at the sampler's door. The sampler needs model, positive, and negative as three inputs; this node gathers them so the connection clutter concentrates in one labeled place.

The inputs and outputs

• model (MODEL) - required. The diffusion model, usually straight from the checkpoint loader.
• positive (CONDITIONING) - required. The encoded positive prompt from CLIPTextEncode.
• negative (CONDITIONING) - required. The encoded negative prompt.

All three come back out under the same names, model, positive, and negative, and all three feed the KSampler's inputs directly. This makes it the natural companion to the Checkpoint Bus on the other end: load the checkpoint into one bus at the start, encode your prompts, bundle the encoded results plus model into this bus near the sampler, and the middle of the graph stays clean.

One distinction worth keeping straight: if your workflow needs to pass the raw prompt text around - say, to several encoders or a wildcard processor - grab the Prompt Bus instead. This node operates on already-encoded conditioning, and it won't help you if your strings haven't been encoded yet. It's the "last mile" of the prompt pipeline, not the entry point.

Installing it

Same pack as all the bus nodes. ComfyUI Manager: search "comfyui-bus-plugin", Install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/rhdunn/comfyui-bus-plugin

then restart. Pure Python with no dependencies - no requirements.txt, no models to download.

Common issues

The common stumble is wiring raw strings into the conditioning ports here - they won't fit, because this bus only accepts CONDITIONING. If you're feeding it prompt text, you've grabbed the wrong bus (you want the Prompt Bus, then a CLIPTextEncode). Also, because the node is a pure pass-through, whatever you connect is what the sampler gets: a mismatched model and conditioning (say, an SDXL conditioning into an SD1.5 model) will pass through unchallenged and produce garbage. The bus keeps your graph tidy; it won't keep you from your own wiring mistakes.