Prompt Text (WFS)

It's a two-string passthrough, so let's not oversell it. WFS_PromptText holds a positive and a negative prompt as plain text on your canvas and outputs both as STRING. That's the whole job - the entire execute() is literally return (positive, negative). No CLIP encoding, no model loading, no cleverness.

So why does it exist? Workflow Studio's Library side panel lets you save prompt presets (positive + negative pairs) and drag them straight onto the ComfyUI canvas. This is the node that gets created when you do - pre-filled with the saved text. It's how the studio's preset system lands inside your graph without you copy-pasting text around or leaving your prompts parked in a file you'll forget. If you never use the Library, you may never touch it, and that's fine; it's a convenience node, not a required one.

Where it fits in a real workflow: a bare string can't be sampled, so you wire the two outputs into a CLIPTextEncode node's text inputs and let that do the encoding. The quiet win is that the prompt survives as an actual node on the graph - and Workflow Studio's Metadata tab knows it: its prompt-extraction chain recognizes WFS_PromptText, so when you regenerate from a saved PNG's embedded workflow, your positive and negative text come back intact instead of being guessed from whatever CLIPTextEncode happened to be nearby.

The inputs (both required, both multiline textareas):

• positive - what you want
• negative - what you don't

The outputs mirror them exactly: positive (STRING) and negative (STRING).

Install - it ships with ComfyUI-Workflow-Studio, so it's the pack install: ComfyUI Manager, search Workflow Studio, or

cd ComfyUI/custom_nodes
git clone https://github.com/ketle-man/ComfyUI-Workflow-Studio.git

then restart. No model files, no extra pip packages for this node; the repo's requirements.txt (onnxruntime, piexif) is only for the Tagger tab.

There's almost nothing to break here - one trap and one gotcha worth knowing. The trap: expecting it to encode. It won't; whatever connects downstream does. The gotcha is model-dependent: on guidance-distilled models (Flux, Z-Image Turbo and friends at CFG 1) negative prompts are basically inert because there's no unconditional pass to steer, so that negative output may do nothing no matter what you type. That's the model, not the node - restate constraints in the positive prompt and move on. For everything else, it's the most dependable node in the pack: it does one thing and does it with zero moving parts.