🐟Unified Prompt Generator

If you've ever built a character prompt by hand - typing out hair color, skin tone, outfit, pose, lighting, camera angle, all comma-separated and hoping you didn't forget one - UnifiedPromptGenerator is that whole process turned into a node full of dropdowns. Pick from curated presets for every attribute, and it hands you back one finished prompt string, ready to plug into your text encoder. No more re-typing "masterpiece, best quality" for the hundredth time.

How it works

There's no cleverness under the hood - it's a big lookup table. Every field is an enum: you pick an option like wavy.beach: rose gold beach waves with dark roots, and the node substitutes in the English description behind it. Set enough fields, hit run, and it concatenates all your selections (skipping anything left on none) into a single final_prompt string. It covers a lot of ground: physical description (age/gender, ethnicity, hair, skin, eyes, lips, body type), outfit (top, bottom, shoes, hat, jewelry, glasses, handbag), and the photography side - pose, expression, scene, composition, camera view, angle, focus, style, and a trailing quality-tag boilerplate. That's 23 required fields in total, which sounds like a lot until you notice most of them already have sensible defaults, so you can leave half of them alone.

The inputs and outputs that matter

You won't touch all 23 fields every time. The ones worth dialing in:

• age_gender, hair, body - set your character's identity. These default to a specific look rather than "none," so if you don't touch them you'll get the defaults every run.
• top / bottom / shoes / hat / jewelry / glasses / handbag - all default to none, meaning outfit is left unspecified unless you pick something.
• pose, expression, scene_preset - these are where most of the creative variation lives; pose and scene_preset default to none too.
• quality - a toggle between nothing and a fixed "masterpiece, best quality, highly detailed..." boilerplate tag string appended at the end.

The single output, final_prompt (STRING), wires straight into a CLIP Text Encode node as your positive prompt.

How to install it

Search ComfyUI-QHNodes in ComfyUI Manager, or clone it by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/liuqianhonga/ComfyUI-QHNodes.git

Restart ComfyUI and the node shows up under the 🐟QHNodes category. No extra models or heavy dependencies - this is a pure string-assembly node, so there's nothing to download. Worth knowing going in: this is a small, personal-use pack (the author says as much in the README - built for their own workflows, shared in case it's useful to someone else) rather than a widely-adopted one, so don't expect a big community thread if you hit something odd. You're mostly on your own with the README and the node itself.

Common issues & troubleshooting

The prompt is way too long and half of it seems ignored. With every field filled in, final_prompt can run to several hundred words. If you're on an SD 1.5/SDXL-family checkpoint, the CLIP text encoder truncates at 77 tokens per chunk - anything past that point is silently dropped, so your carefully chosen shoes and jewelry might never reach the model. This node is a much better match for checkpoints with a full-sentence text encoder (Flux, and most 2026-era models), which actually read the whole thing as an instruction rather than a bag of tokens.

The quality boilerplate doesn't seem to do anything. That's expected on newer, LLM-encoded checkpoints - masterpiece, best quality, 8k uhd is vocabulary trained into SD 1.5/SDXL-style CLIP encoders, and it's essentially inert on models with a different text encoder. Turn it off (set quality to none) rather than assuming it's broken.

Output looks like a mash of two languages. Every dropdown label shows an English phrase followed by a Chinese translation in parentheses - that's just the picker UI. Only the English half before the parenthesis is meant to end up in the actual prompt; if you're seeing Chinese characters in your final image's prompt overlay, something upstream (a text-display node, a screenshot) is showing you the raw enum label rather than the resolved string.