Portrait Fashion Prompt

You know the grind: you want a character in a doctor's coat, leather jacket, and high-waisted jeans, and you spend ten minutes fighting the tokenizer over whether it's "jeans" or "denim pants" before giving up and writing "cute outfit". Portrait Fashion Prompt exists to end that. It's a giant dropdown form - 748 kinds of clothes, 363 colors, plus hats, shoes, gloves, and a dozen jewelry/decoration slots - and it turns whatever you pick into a proper weighted prompt. It's from ComfyUI-PromptWrapper, a Chinese-built pack that is basically unknown in the English-speaking community (zero exact matches in the reddit corpus), and that's a shame, because this is its most popular node for a reason.

How it works

The node is a front end over bundled datasets - JSONL files of clothing terms in the pack's assets/ folder, in both Chinese and English. Each dropdown has a -- (omit this) and a random 🎲 option. Pick random and the node does seed % dataset_length to choose an entry, so the same seed gives you the same outfit every time. Weights work the way they do in a normal weighted prompt: weight 1 means the bare text, anything else becomes (text:1.35).

The inputs that matter

You don't touch all 45 of them. Realistically:

• clothes and clothes_color - the two that carry the image. Leave the rest at --.
• clothes_weight, clothes_color_weight - the sliders (0–2) that decide how hard the model tries. 0 omits the slot entirely.
• seed - reproduces any random pick.
• language - Chinese or English; this swaps the dataset, so you can generate bilingual prompts.
• preset_prompt and the optional input_prompt - both get prepended to whatever the node builds, handy for a fixed subject line like "1girl, standing".

The single text output wires straight into a CLIP Text Encode node.

Install

ComfyUI Manager → Custom Nodes Manager → search ComfyUI-PromptWrapper → Install, then restart ComfyUI. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/clouddreamfly/ComfyUI-PromptWrapper

The builder nodes need nothing extra - the datasets ship with the pack. No models, no API keys.

Where people get burned

Two things. First, the (word:weight) syntax the node emits is a CLIP-encoder convention: it works great on SD 1.5/SDXL-lineage checkpoints like Illustrious or Pony, but 2026's LLM-encoded models (Flux with T5, Z-Image, Anima) silently discard those weights. On those, this node is just a fancy tag picker. Second, the default language is Chinese - switch it to English if you don't want 医生的白大褂 showing up in your prompt, because the "English" dataset is genuinely translated and reads better to most models.

It's not a workflow engine and it won't compose a scene for you. It's a fast, structured way to describe clothing, and for character work it beats typing tags by hand.