China Attire

Ask a base model for a qipao and you often get a satin party dress with a slit and a dragon printed on it. The China Attire node from ComfyUI Globetrotter Nodes is here to make sure the prompt actually names the real garment and describes it properly - the collar, the side slit, the cut - so the model has a fighting chance of rendering it right. It's a prompt-generator node: pick your options, get a finished prose prompt out the other end, wire it into your positive prompt.

What it is

One of the country nodes in manifestations/comfyui-globetrotter, built the same way as the rest: a factory generates the node class at load time from JSON files in data/, and every option you see in the dropdowns comes from data rather than Python code. The honest framing: China is one of the thinner entries in the pack. There's no regional appearance data (region is none/random/Unspecified only), no country-specific poses, and no landmark settings - the setting list is the generic shared one. What it does have is a solid core of authentic upper-body garments, and that's what you're here for.

How it works

On run, the node assembles a natural-language prompt in a fixed order: subject → appearance → emotion → attire descriptions → pose → setting → atmosphere → composition → quality keywords. Attire items are gender-filtered and each carries a full descriptive sentence, which is the whole trick - the garment name plus what it looks like. Pick Upper_Body: Qipao (Cheongsam) and the output doesn't say "qipao," it says the dress with its Mandarin collar and slit, described in enough words for the model to build on.

The inputs that matter

• gender and age set the subject; attire options filter around them.
• Upper_Body is the headline dropdown: Qipao (Cheongsam), Hanfu, Tangzhuang (Tang Suit).
• Head: Phoenix Crown (Fengguan), Mianfu Crown, Guan - the crown options are the genuinely fun part if you want a ceremonial look.
• Legs: Ruqun, Kun, Lotus Shoes. Waist: Yudai.
• randomization (off/light/moderate/full) and seed control reproducibility; the author's tooltip is unambiguous: same seed + same settings = consistent outputs. Full randomization deliberately ignores the seed.
• lora_trigger, custom_prompt, and experimental_llm_rewrite round it out. The last one needs transformers installed and downloads distilgpt2 on first use - skip it unless you specifically want that.

Output

One prompt STRING. Feed it to a CLIP Text Encode (Prompt) node's text input as the positive prompt, then to KSampler.

Install

cd /path/to/ComfyUI/custom_nodes
git clone https://github.com/manifestations/comfyui-globetrotter
cd comfyui-globetrotter
pip install -r requirements.txt   # optional

Restart ComfyUI (you'll see "Globetrotter: Loaded N node classes." in the console), or use ComfyUI Manager → search "Globetrotter." No model files to download for the node itself.

Troubleshooting

• Garment choice has no visible effect: check you picked an actual option rather than leaving it on "none" - every dropdown defaults to none/random and "none" means "skip this entirely."
• Fewer attire options than a neighboring country: expected. China's data is thinner than India's or Indonesia's; this is a difference in the pack's data coverage, not something you broke.
• experimental_llm_rewrite does nothing: transformers isn't present, and the node quietly skips instead of erroring.
• Small pack, no community write-ups to lean on - when in doubt, read the "Could not load …" warnings in the ComfyUI console.