Action

If every character you generate looks like a mannequin waiting for a bus, this is the node you're missing. The Action node is the Prompt Builder pack's "what is the subject doing" section - pose, gesture, and movement in a set of dropdowns, so you don't have to remember that the Danbooru tag for a character resting weight on one hip is some hyphenated phrase you'll get wrong anyway.

It exists because motion tags are the difference between a stiff hero shot and something that reads as alive. Most anime-SDXL checkpoints (Pony, Illustrious, NoobAI) were trained on Danbooru's tag vocabulary, so a curated list of the pose tags those models actually understand is genuinely useful.

How it works

Action is one of the pack's simple "section" nodes. It reads its dropdown options from presets.json, joins whatever you selected with commas, and outputs a single string. Anything left on None is skipped entirely. There's no magic here - just six dropdowns and a comma join, which is exactly why it's so cheap to use.

The six inputs break down cleanly:

• pose and pose_2 - 63 options each, and having two slots is the pack's way of letting you layer. Pick a base pose in pose ("sitting", "leaning") and a modifier in pose_2 ("lying down", "standing with weight shift") if the situation calls for it. Otherwise leave pose_2 on None.
• posture - the vibe of the body: relaxed, tense, dynamic, confident, shy. Small dropdown, surprisingly large effect.
• action - 44 verbs: jumping, casting spell, fighting, dancing, reaching out, looking back.
• gesture - what the hands are doing: hand on hip, peace sign, arms crossed, finger gun. This is where you un-stiffen a character.
• interaction - eye contact and who they're engaging with: looking at viewer, looking away, holding weapon, hugging another person.

The single output is action - a string that wires straight into the action input on the pack's Combine Prompt node.

Install

No dependencies, no model downloads - this pack is pure Python string work. Either clone it into custom_nodes or use ComfyUI Manager (search "Prompt Builder"), then restart ComfyUI.

cd ComfyUI/custom_nodes
git clone https://github.com/zinigo-creations/comfyui-prompt-builder

Common issues

The trap with Action is over-stuffing. Pick two or three of the six dropdowns and you have a lively character; set all six with conflicting verbs - "running" plus "sitting" plus "looking back" - and the model will mush them into something nobody asked for. The pack's own README says simpler prompts often produce better results, and that warning is aimed squarely at this node. Also worth knowing: these are Danbooru-derived tags, so they only speak to models trained on that vocabulary. On a 2026 LLM-encoded model the words still say something, but you're trading away the exactness this pack was built around.