APNext Poses

PosesPromptNode (APNext Poses) is the offline builder for how your subject is positioned and framed. It appends pose and shot-type phrases to your base prompt from two curated banks - body poses and portrait framings - with no API, no key, no model. It's a small, focused member of the pack's Auto Prompter family, and it earns its keep when you're batching variations of a character and want the pose to change each time instead of getting the same headshot on repeat.

How it works

The node holds two lists and stitches your picks onto the prompt with the separator (default ,). Both dropdowns offer None, Random, Multiple Random, or a specific pick:

• poses (150+ options) - body positions and gestures: "crouching," "blowing a kiss," "dancing or jumping in mid-air with a joyful expression," "holding an object in one hand, such as a flower or book."
• portrait_poses (~28) - framing and shot type: "close-up," "3/4 shot," "full body shot," "environmental portrait," "head tilted to the side," "arms crossed."

seed makes random rolls reproducible; attributes adds extra descriptors. Two outputs - prompt (base plus the picks) and random (a randomized selection) - into your text encode.

The clean split between poses (what the body does) and portrait_poses (how the camera frames it) is the useful part: pair "arms crossed" with "3/4 shot" and you've specified both the subject and the shot in two clicks.

Poses vs. real pose control

Worth being straight about scope. This node adds pose words to a text prompt - "crouching," "full body shot." That nudges the model, and on high-adherence modern models it nudges pretty reliably. But it is not pose control. If you need the subject in an exact, specific pose - a precise limb arrangement, a matched reference - that's a job for ControlNet with an OpenPose skeleton, which conditions the image on an actual pose map rather than hoping the word lands. Think of PosesPromptNode as the fast, loose way to get pose variety; ControlNet is the way to get a pose exactly.

Installing it

ComfyUI Manager: search comfyui_dagthomas, install, restart. Or clone and pip install -r requirements.txt. This node needs none of the pack's LLM dependencies, so it runs regardless of the API nodes.

Common issues

The pose is close but not exact. That's the ceiling of prompt-based posing. For an exact pose, drive it with ControlNet OpenPose instead of, or alongside, this node.

Framing and pose fight each other. "Close-up" plus "full body shot" is a contradiction; the model picks one or averages badly. Choose one framing from portrait_poses.

Nothing changed. Both dropdowns at None append nothing - set at least one to a value or to Random.