🐳Prompt解析

Prompt Parser (🐳Prompt解析) is a tiny text-splitting utility that exists because of a formatting convention: when a structured prompt or template looks like

Category: [clothing]
Color: [red]
Instruction: [Put a clothing on the model.]

you want each [bracketed] chunk on its own wire, not jammed into one string. This node extracts every bracketed piece and gives you each one as a separate STRING output. Its default input text is literally that three-line example - the author ships it self-documenting, which tells you exactly what it's for.

The mechanism is one regex: it finds everything inside [...] (non-greedy, so nested-looking brackets don't over-consume), strips the surrounding whitespace, and returns them in order. The output_count input (1-100, default 3) controls how many output slots get filled: if you extracted fewer fields than output_count, the leftover outputs come back as empty strings; if more, the extras are dropped. That padding is why the node always gives you a stable number of outputs to wire up even when your input text varies.

The node declares 100 STRING outputs (输出_1 through 输出_100), but the pack's JavaScript hides the ones you're not using - you only see as many as output_count asks for. So don't be intimidated by the schema; on the canvas it just looks like a node with however many wires you set.

Where it earns its place

This pattern - structured bracketed templates - is exactly what this pack's other prompt and selector nodes produce. Pose Selector, Clothing Selector, and the Wan2.2/Qwen prompt templates in the pack all work by building structured prompt text with fields you'd want to separate before they hit different parts of your graph (an instruction field into one conditioning, a category field into another, a color field into a color-using node). Text Format Parser is the adapter that unpacks them.

It's also handy if you have an LLM node or API upstream that returns bracketed output and you want per-field routing. Feed the raw response in, set output_count to your known field count, and each field becomes a first-class wire.

Honest limits: it only extracts bracketed text - if your fields use {} or () or a colon-separated scheme, this won't see them. And it's purely positional; if the template is missing a field, your downstream node gets an empty string, which may or may not be what you want (empty strings are often fine for text encoders, but they don't silently "skip"). For the bracketed-template world this pack lives in, though, it's the right tool.

Install

Part of Pond Nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/Pondowner857/comfy_Pond_Nodes
cd comfy_Pond_Nodes
pip install -r requirements.txt

Restart after (or Manager → "comfy_Pond_Nodes"). No models, no optional deps - pure regex plus a small JS file for the dynamic outputs. Standard pack caveat: the README warns of console spam if comfyui_HiDream-Sampler is installed alongside.