Load YOLO Model

This is the boring half of the pack, and boring is a compliment. YOLOModelLoader ("Load YOLO Model (For SDPose)") exists for one reason: to hand BBoxYOLO a loaded ultralytics YOLO model so it can find the people that ComfyUI's official SDPose keypoint extractor won't find on its own. You will never spend more than ten seconds on this node - the entire UI is a single dropdown.

How it works

The loader reads whatever .pt weight files it finds in ComfyUI/models/yolo - a model folder the pack registers on install - and constructs an ultralytics.YOLO object from the one you pick. That object comes out the YOLO_MODEL output, which is the only thing BBoxYOLO's yolo_model input accepts. It's a thin wrapper around YOLO(model_path), nothing clever, and that's fine: the clever part of the pipeline is downstream.

The one real subtlety: this pack does not download weights for you. The README is a single line plus a workflow screenshot; there's no model link, no auto-download hook. Unlike some one-click packs, install does not make it work. You supply the file.

Getting a model file

Put any COCO-trained YOLO weight in ComfyUI/models/yolo (create the folder if it's not there yet). A tiny yolov8n.pt (~6 MB) is genuinely enough for person detection - this is a localization task, not a segmentation contest - and it runs noticeably faster than the s or m sizes. yolo11n.pt works too if you'd rather run the newer arch.

cd ComfyUI/models/yolo
wget https://github.com/ultralytics/assets/releases/download/v0.0.0/yolov8n.pt

Restart ComfyUI (or hit Refresh), and the loader's dropdown will list it. If the dropdown is empty, the pack is installed but you skipped this step - the folder is empty, so get_filename_list("yolo") has nothing to show.

Install

ComfyUI Manager: search "ComfyUI_YOLO_For_Multi_SDPose_Detection" and install. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/judian17/ComfyUI_YOLO_For_Multi_SDPose_Detection

then restart. The requirements.txt is just ultralytics and numpy, but ultralytics pulls in torch, torchvision, and OpenCV, so let Manager finish the pip install before you judge it broken - a fresh install that hasn't restarted after installing deps will throw the pack's "ultralytics library not found" warning at startup and the loader will refuse to load anything. After a proper restart you're fine.

Gotchas worth knowing

• It's not a pose model. It detects bounding boxes; SDPose does the keypoints. Don't expect keypoints out of this node - the pack's division of labor is loader → detector → official SDPose.
• Custom-class weights break it silently. The detector assumes class 0 = person. Standard COCO weights (v8/v11) are safe; anything you fine-tuned on other classes will find nothing.
• Watch the dropdown on fresh clones. The pack registers the yolo model folder at import time, so it shows up even before you've added files - an empty folder just means an empty list.

Judian17 is a small but real author in this space - same person behind ComfyUI-SDPose-OOD, which people actually reach for when DWPose gives black output on cartoon frames. This loader isn't that pack; it's the mundane plumbing that makes the multi-person SDPose workflow hold together. Mundane is the point.