VNCCS Pose Generator

This node draws a skeleton - the same 18-point body layout (nose, neck, both shoulders, elbows, wrists, hips, knees, ankles, eyes, ears) that OpenPose has used for years - onto a plain canvas, one skeleton per pose. What it doesn't do, and this is worth knowing before you go looking for a ControlNet checkpoint slot on this node, is feed straight into a ControlNet.

VNCCS's Pose Studio is described by the pack's own author as one of the project's main nodes as of the 3.0 rewrite, letting a character "strike absolutely any pose." That capability comes from Qwen-Image-Edit 2511, the model VNCCS's newer pipeline runs on, with dedicated LoRAs handling the posing itself rather than a traditional pose ControlNet. That's a broader shift the whole ecosystem has been making - reference-image posing via an edit-model LoRA instead of extracting a skeleton and running it through a ControlNet checkpoint (the character-consistency writeup covers this move in detail, and Qwen-Image-Edit's own entry notes it consumes pose maps as reference input rather than through a separate ControlNet). Pose Generator's job in that world is producing the reference skeleton image the pose LoRA reads, not a ControlNet conditioning map.

The inputs that matter

• pose_data - a large JSON blob defining one or more poses as sets of body keypoints, each a canvas-space [x, y] coordinate. You won't hand-type this; it comes from wherever Pose Studio's interactive pose editor writes it, letting you drag joints around visually rather than editing coordinates directly.
• line_thickness (default 3, range 1–10) - how thick the rendered skeleton lines are.
• safe_zone (default 100, range 0–100) - a margin that keeps the rendered skeleton from touching the canvas edges, presumably so nothing gets clipped when the pose is used as a reference downstream.

One output, openpose_grid - an image, note the name: it's built to lay out multiple poses in a grid on one canvas (the default pose_data sample carries a dozen separate pose entries on a single 512×1536 canvas), matching how VNCCS generates a character's whole pose set in fewer, batched passes rather than one generation per pose.

How to install it

Bundled with the main pack, but with one extra step worth knowing about: the README specifically calls out that Pose Studio is downloaded from AHEKOT's companion project, VNCCS-Utils, so don't skip installing that alongside the main pack.

• ComfyUI Manager - search VNCCS - Visual Novel Character Creation Suite, install, restart, then click "Install missing custom nodes" so the VNCCS-Utils companion pack (and this node with it) actually gets pulled in.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/AHEKOT/ComfyUI_VNCCS.git, then cd ComfyUI_VNCCS_Utils && pip install -r requirements.txt, restart, and run "Install missing custom nodes" in Comfy Manager once.

Common issues

Poses come out looking anatomically wrong on your character even though the skeleton looks right. Set the character's body proportions - age, height, body type - in Pose Studio before posing rather than after. The README is explicit that age carries over automatically from Character Creator, but height and body type need setting manually, and getting them right materially affects how well the pose transfers onto the actual generation.

You imported a pose from a reference picture and it doesn't match what you expected. Pose Studio's Import feature extracts a pose from any image you feed it, which is convenient but only as accurate as the pose-extraction step itself - a reference photo with an ambiguous or partially occluded pose (a hand behind the back, limbs overlapping) can produce a skeleton that misreads the original. Check the rendered openpose_grid output before running a full generation off it.

Skeleton looks clipped at the edges. Raise safe_zone - it exists specifically to pull the drawn skeleton in from the canvas border, and a value that's too low for an especially extended pose (arms fully raised, wide stance) can still let joints run to the edge.