ComfyUI-Pose-Loader

Loads pose packs from a directory and provides them as an image batch. Allows you to batch-generate images from pose packs.

Killer feature: you don't have to unzip the pose packs, just put them in a directory and go. Or let your CivitAI downloader put them there automatically.

Other reasons that this exists:

• reads OpenPose-annotated images and depth maps directly from zip files
• filters out preview images and other junk, so your ControlNet only sees the poses
• turns your collection of pose packs and misc unsorted OpenPose pngs into a single source of poses
• basically makes it easy to use your giant pile of poses with ComfyUI

Installation

cd ComfyUI/custom_nodes && git clone https://github.com/ceruleandeep/ComfyUI-Pose-Loader.git

Usage

Download pose packs from CivitAI or whatever and put them in a directory. By default this is ComfyUI/custom_nodes/ComfyUI-Pose-Loader/models/poses.

Add a PLLoadImagesFromDirBatch node to your graph. Connect the images output to a ControlNetApply node.

Set the following parameters to your liking:

• directory: Subdirectory of your poses directory to load from, or empty for the base directory.
• image_filter: Attempts to load only OpenPose annotations or depth maps, so the right poses are sent to the ControlNet.
• image_load_cap: Maximum number of images to load from the directory. May return less!
• start_index: Index of the first image to load. Useful for loading bits of a pose pack, or when loading multiple pose packs at once.
• recursive: Whether to descend into subdirectories of the selected directory.

Configuration

You probably have a different directory for your pose packs. Configure it in extra_model_paths.yaml.

Example configuration to load poses from AUTOMATIC1111:

a111:
    base_path: /home/user/stable-diffusion-webui/models/
    poses: Poses

Returns

• images: Batch of images, shaped according to the size of the first image loaded.
• masks: Batch of masks, shaped however the masks are shaped.
• count: Number of images loaded. Feed this into your Efficient Loader or whatever, so you can e.g. generate one image per pose.
• width: Width of the images. Theoretically you could use this to generate images with the same size as the pose maps.
• height: Height of the images.

Credits

Batching code adapted from ComfyUI-Inspire-Pack