Prompt Builder (CLIP) — ComfyUI custom node

Stop scrolling through one giant wall of text. Prompt Builder lets you split your prompt into named, reorderable fields (Character, Face, Clothing, Background, Lighting…) and merges them into a single prompt, encoded straight to CONDITIONING.

Features

• Dynamic fields – add as many as you need with one click

• Custom names – label each field however you like (the names are for you only and are not added to the prompt)

• Reorder – move fields up/down with ▲ / ▼; the order defines the final prompt order

• Toggle on/off – disable a field with a checkbox without deleting its text

• Resizable text boxes – the height you set is remembered and never resets

• Saved with the workflow – fields, text and sizes are restored when you reopen it

• Custom separator – default , , change it to anything you want

• Clean merging – empty fields are skipped, stray commas at field edges are trimmed

Inputs / Outputs

Type Name Description Input clip CLIP model Output CONDITIONING Encoded prompt, plug into your sampler Output prompt Final merged prompt as a string (handy for debugging)

Installation

1. Go to ComfyUI/custom_nodes/

2. Create a folder named prompt_builder

3. Put the files inside so it looks like this:

ComfyUI/custom_nodes/prompt_builder/
├── __init__.py
└── js/
    └── prompt_builder.js

1. Restart ComfyUI and hard-refresh the page (Ctrl+F5)

2. Double-click the canvas, search Prompt Builder, or find it under conditioning/prompt

Usage

1. Connect your CLIP to the node

2. Click + Add field, name it and write your text

3. Reorder with ▲ / ▼ until the prompt reads the way you want

4. Connect CONDITIONING to your sampler (positive or negative, just use a second node for the other one)

Example

Field Text Character 1girl, red hair, long hair Face detailed face, green eyes, smile Clothing black leather jacket, white t-shirt

Result:

1girl, red hair, long hair, detailed face, green eyes, smile, black leather jacket, white t-shirt

Notes

• No extra dependencies, pure ComfyUI.

• Uses the standard clip.tokenize + encode_from_tokens_scheduled, so it works with any CLIP-based model.

• If an existing node looks wrong after an update, delete it and add a fresh one.

Feedback and suggestions are welcome in the comments!