Skeleton styles

A pose-conditioned model follows the kind of skeleton drawing it was trained on, so poseorbit draws two.

Style	Drawing	For
dwpose	rtmlib's COCO-WholeBody drawing: thin coloured lines and dots for the body, feet, hands and face	DWPose-trained adapters, such as Anima-Control-Pose
openpose	The 18 OpenPose body points, thick translucent limbs whose width grows with the canvas	OpenPose ControlNets, such as xinsir/controlnet-openpose-sdxl-1.0

Both are on black, at the output's size.

dwpose

Drawn by rtmlib's draw_skeleton with the same keypoint threshold (0.3), from the same DWPose detector skeleton-conditioned models are commonly trained on, so a skeleton looks like their training data. Hands and face are included, which helps in close-ups and for which way the head faces.

openpose

Follows the drawing on the model card of xinsir/controlnet-openpose-sdxl-1.0, which replaced controlnet_aux's thinner one: limb and joint thickness steps up with the canvas's longer side (one size below 500 px, two below 1000, then one more per 1000 px). Body only, as that model was trained. In a close-up few body points remain; check PoseResult.joints_in_frame.

Measured

Generated with and without the skeleton, same seed, 832×1216; PCK@0.1 of the re-detected output against the skeleton (1.0: every point within 10% of the diagonal of where the skeleton put it).

Model	Pose	Without	With
Illustrious XL 2.0 + xinsir ControlNet (openpose)	walking, front	0.76	0.94
same	walking, turned yaw 45° / yaw −60° pitch 15°	—	0.94 / 0.88
same	upper body ×2 / face ×5	0.02 / 0.48	1.00 / 0.99
Anima Base 1.0 + Anima-Control-Pose (dwpose)	walking, front	0.06	0.71
same	walking, turned yaw 45°	—	0.94