🐳身体部位选择器

Regional work in ComfyUI - ADetailer passes, regional prompting, per-part inpainting - all has the same annoying preamble: typing out "face, hands, hair" or "left arm, right leg" into a dozen different prompt boxes, and retyping it when you change the subject. Pond_c_e (display name "🐳身体部位选择器," body part selector) from 🐳 Pond Nodes (comfy_Pond_Nodes) turns that into checkboxes: tick the parts you care about, get back a comma-separated English string you can drop into any prompt input.

The class name is the opposite of descriptive - Pond_c_e tells you nothing - but the node itself is dead simple, and that's the appeal.

How it works

Twenty-five boolean toggles, each mapping a Chinese label to an English token. Tick 脸 (face), 手 (hands), and 头发 (hair), run it, and the output is face, hands, hair. Untick everything and you get an empty string. No models, no inference - it's a lookup table wearing a checkbox UI. The part list is thoughtful: the usual anatomy (face, hands, eyes, mouth, nose, ears, body, arms, legs) plus a solid clothing set (top, jacket, T-shirt, shirt, pants, shoes, skirt, dress, tie, belt, backpack, scarf, glasses), so it doubles as a wardrobe-tag picker.

One output: STRING - the assembled comma-separated English tags.

Where it fits

The obvious use is regional detailers: wire the string into an ADetailer-style node's prompt so the face pass says "face," the hand pass says "hands." It also plays well with regional prompting and with ControlNet pose work where you want the prompt to name the parts the conditioning is controlling. Since the output is plain text, it composes with anything - concatenate two instances for multi-character scenes, or prepend your main subject.

Installing it

Standard pack install. Manager search comfy_Pond_Nodes, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Pondowner857/comfy_Pond_Nodes
cd comfy_Pond_Nodes
pip install -r requirements.txt

Restart, and it's under 🐳Pond/选择器. Standard library only - skip the pack's heavy dependencies if you only want the selectors. (Pack-wide footnote: the README flags a console-spam conflict with comfyui_HiDream-Sampler.)

The gotcha

It's a tag builder, not a quality guarantee. The English output is intentionally plain - face, left arm, T-shirt - which is exactly right for CLIP and tag-trained checkpoints, but if you're on a natural-language model (Flux, Qwen-based) you may want to wrap the result in a sentence. And one footgun: the toggles are all false by default, so a fresh node silently outputs an empty string and downstream nodes just run without the region. If your detailer suddenly stops targeting hands, check whether the checkbox actually got ticked.