Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.comfy.org/llms.txt

Use this file to discover all available pages before exploring further.

This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub
The SDPoseKeypointExtractor node detects human pose keypoints from input images using the SDPose model. It can process full images or specific regions defined by bounding boxes and outputs the detected keypoints in the OpenPose format, which includes the coordinates for each person and a confidence score for each keypoint.

​
Inputs

ParameterData TypeRequiredRangeDescription
modelMODELYes-The SDPose model used for keypoint detection. Must be a model with a heatmap_head attribute, specifically from the SDPose repository.
vaeVAEYes-The VAE model used to encode the input images into the latent space for processing.
imageIMAGEYes-The input image or batch of images from which to extract pose keypoints.
batch_sizeINTNo1 to 10000The number of images to process at once when running in full-image mode (i.e., when bboxes is not provided). This can speed up processing. (default: 16)
bboxesBOUNDINGBOXNo-Optional bounding boxes for more accurate detections. Required for multi-person detection. If provided, the node will extract keypoints from each specified region.
Parameter Constraints:
  • The model input must be a specific SDPose model. If the provided model does not have a heatmap_head attribute, the node will raise an error.
  • The node operates in two distinct modes based on the bboxes input:
    1. Bounding Box Mode: When bboxes is provided, it processes each specified region individually. This is required for detecting multiple people in a single image.
    2. Full-Image Mode: When bboxes is not provided, it processes the entire image as a batch. The batch_size parameter only applies in this mode.
  • When bboxes is provided, it must be a list of bounding boxes. If there are fewer bounding box lists than images, the last list is reused for remaining images.
  • Bounding boxes with zero width or height (e.g., x2 <= x1 or y2 <= y1) are skipped.

​
Outputs

Output NameData TypeDescription
keypointsPOSE_KEYPOINTKeypoints in OpenPose frame format (canvas_width, canvas_height, people). The output contains the detected persons, each with an array of keypoint coordinates (x, y) and their corresponding confidence scores.
Source fingerprint (SHA-256): 5e199e9d08a46d934c6d1f4b1a411f0ca95eb5e7f85ec39f59a058c9be598e9f