CreaPrompt node 1

Of the four numbered CreaPrompt nodes, this is the one people actually find - it's the most-visited node in the pack, full stop. And there's a reason: node 1 is the woman and fashion configuration. Its dropdowns are the categories that make portraits interesting - who she is, what she's wearing, her hair, the shot, the place - and nothing else. If you want a machine that generates endless believable portrait prompts, this is the one to drop in.

Each numbered node is just the same random-prompt engine pointed at a different CSV folder, and this one reads csv1: Woman, Celebrity_Woman, Fashion, Shotstyle, Haircuts_Thomas_Buyle, Haircolors_Thomas_Buyle, Woman_Dress_Malapris_PJ, and Places. That's it, eight categories, and that tightness is the point. The complete node's 47 dropdowns are great for wandering; this one is for sitting still and generating a hundred portrait prompts that stay on theme.

The inputs that matter

Every category defaults to disabled - a category contributes nothing until you touch it. The option that matters is 🎲random, which picks a random line from that category's CSV on every run. The usual recipe:

• Woman and Fashion on 🎲random - the subject and the outfit, which are the heart of the prompt.
• Haircuts_Thomas_Buyle and Haircolors_Thomas_Buyle on 🎲random - 58 haircuts and 57 colors, and they're actual descriptive lines ("Balayage", "Blunt bob"), not just tags.
• Shotstyle and Places on 🎲random - the framing and the backdrop.
• Celebrity_Woman - the wildcard. 950 celebrity names, from A.J. Cook to Adriana Lima. It's a "style by this person" play, not a likeness guarantee; models handle it accordingly.

Then the three optional inputs on every node in this pack:

• Prompt_count (1–1000) - emit that many random prompts, one per line.
• CreaPrompt_Collection - flip to enabled and the categories are ignored; it pulls complete prompts from collection.txt instead.
• seed - lock it and the same picks repeat.

Outputs: prompt (STRING) into a CLIP Text Encode, seed (INT) back into the KSampler for reproducibility. Standard wiring, nothing exotic.

Installing it

Same as the whole pack - ComfyUI Manager (search "CreaPrompt") or:

cd ComfyUI/custom_nodes
git clone https://github.com/tritant/ComfyUI_CreaPrompt

Restart, done. No pip installs, no API keys, no model downloads - this is the zero-friction version of the pack. The author's recommended companion checkpoint, CreaPrompt_Ultimate (CivitAI 383364), is an SDXL Lightning/Hyper model built to run at low CFG and pairs naturally with these prompts, but any SDXL checkpoint works.

Common issues

• New categories don't show up. Drop a new CSV into ComfyUI_CreaPrompt/csv1/ and re-add the node (or reload the workflow) - the dropdowns are built when the node loads.
• Never delete collection.txt from csv1 - the node errors without it.
• Celebrity prompts can flop. 950 names is a lot of luck; some names your checkpoint has never seen and will render as generic pretty faces. That's the model, not the node - treat it as a lottery, fix a seed when you hit one that works.
• Weights are SDXL-only. The (word:1.2) fragments in the stock CSVs are inert on LLM-encoded models like Flux or Z-Image, which read the parens as literal text. On those models, prefer the natural-language categories and skip the tag soup ones.