ClothingWildcards

A wildcard node with ambitions. Beyond the usual "pick a random line from a text file," ClothingWildcards parses a small [cfg=…, exposure=…, skin=…] block out of the filled-in text and hands those numbers to you as separate float outputs. So one node can randomize an outfit from a wildcard file and tell you what CFG, exposure, and skin-tone weighting this particular pick wants - which makes it a decent fit if you're building a character/outfit dataset where each variation carries its own settings.

How it works

Write your prompt with __wildcardname__ tokens. The node looks for matching .txt files inside the pack's own wildcards/ folder (not ComfyUI's global wildcard directory), reads each file, and replaces every __name__ token with a randomly chosen line. Then it scans the populated text for [key=value] pairs and maps cfg, exposure, and skin to float outputs - with defaults of 7.0, 0.0, and 0.0 when a key is absent. The prompt output is the populated text with the attribute block stripped away; full_prompt keeps it.

Inputs and outputs

One input, wildcard_text (multiline). Five outputs:

• full_prompt - populated text including the [cfg=…] block.
• prompt - populated text with the block removed, ready for the CLIP encoder.
• cfg - FLOAT, default 7.
• exposure - FLOAT, default 0.
• skin - FLOAT, default 0.

Wire cfg into your sampler's CFG input and, in a workflow that exposes exposure/skin as controls, the node effectively becomes self-configuring per pick.

Where it bites

The wildcard files are the pack-local wildcards/ directory - that folder only ever contained a placeholder in the repo, so nothing will match until you add your own .txt files. Misspell a wildcard name and the node just prints "未查到该路径" (path not found) and moves on, leaving the __name__ token intact in your prompt - a quiet failure. And this is one of the nodes that vanished in the April 2025 rewrite: fresh installs only get the captioner, so the wildcards/ folder and this node both have to be pulled from the repo's older history.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/438443467/ComfyUI-GPT4V-Image-Captioner

Restart ComfyUI. No models needed. If you only want random-prompt variety without the attribute plumbing, the pack's SimpleWildcards is the simpler sibling - grab this one when per-pick settings matter.