🐳肢体选择器

If Pond's Pond_c_e is the "which parts do you care about" picker, Pond_c_f (display name "🐳肢体选择器," limb selector) is the same idea taken down to the joint level. Where the body selector stops at "left arm," this one splits anatomy into left eye, right eye, left eyebrow, right eyebrow, left ear, right ear, teeth, hands, feet - the granularity you actually want when you're running a per-part detail pass and need the prompt to name the exact limb being re-rendered.

It's from 🐳 Pond Nodes (comfy_Pond_Nodes), sits under 🐳Pond/选择器, and like its sibling it's a checkbox-to-tags lookup with no model involved.

How it works

Twenty toggles, each mapping a Chinese label to an English token: 头发→hair, 左眼→left eye, 右眼→right eye, 左眉→left eyebrow, 鼻子→nose, 嘴巴→mouth, 牙齿→teeth, 左耳→left ear, 右耳→right ear, 左胳膊→left arm, 右胳膊→right arm, 左手→left hand, 右手→right hand, 左腿→left leg, 右腿→right leg, 左脚→left foot, 右脚→right foot, plus 脸部 (face) and 身体 (body) as catch-alls. Tick what you want, run it, get back a comma-separated English string like left hand, right hand, teeth. One STRING output.

The value over Pond_c_e is precisely the lateral symmetry. Hands and feet are the classic weak points of diffusion, and a detail pass that says right hand instead of just hands tells the model which one to fix - that's the difference between an ADetailer run that improves things and one that shuffles pixels.

Where it fits

Same plumbing as Pond_c_e: feed the output into regional detailers, inpainting prompts, or regional conditioning. If you're doing a ControlNet OpenPose workflow, naming the exact limb in the prompt reinforces what the skeleton is already saying. And because it's plain text, you can combine instances - a Pond_c_f for the hands and a Pond_c_e for clothing, concatenated, for a two-part detailer setup.

Installing it

The whole pack installs the same way - 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

Then restart. Standard library only, no heavy dependencies needed for this node. (Pack-wide note: the README warns about console spam if comfyui_HiDream-Sampler is also installed.)

The gotcha

More granularity means more opportunity to misname. The left/right split assumes your image's perspective - "left hand" from the viewer's point of view versus the subject's. If a detail pass keeps hitting the wrong hand, flip which toggle is on before you blame the node. And remember the tokens are plain English tags (left eyebrow), tuned for tag-trained checkpoints; on a natural-language model you'll want to wrap them in a phrase. Also, everything's false by default - an all-unchecked node outputs an empty string and downstream runs just silently continue.