BMAB Detector

This is the raw building block underneath all of BMAB's detailer nodes, exposed on its own. Feed it an image and pick a YOLO model, and it hands you back a mask of whatever that model found - no redraw, no sampling, no bind, just detection. Every one of BMAB's Face/Hand/Person Detailer nodes is essentially this node plus an automatic crop-resample-composite loop bolted on. Reach for this one directly when you want the mask itself - to feed into your own inpaint chain, to combine with another mask, to preview what a detector is actually catching before trusting it to a full detail pass.

The models on offer

The model dropdown lists nine YOLOv8 checkpoints, split into three families:

• Face: face_yolov8n.pt, face_yolov8n_v2.pt, face_yolov8m.pt, face_yolov8s.pt - nano through medium/small variants, trading speed for accuracy as you go up in letter size.
• Hand: hand_yolov8n.pt, hand_yolov8s.pt.
• Person (segmentation): person_yolov8m-seg.pt, person_yolov8n-seg.pt, person_yolov8s-seg.pt - these produce a proper body outline rather than a bounding box, since they're segmentation models, not plain detectors.

These are the same detection weights the wider ecosystem runs - ADetailer in A1111 and Impact Pack's detector nodes in ComfyUI draw from the same Bingsu/adetailer family, so if you've used either of those before, the naming will already be familiar. That shared lineage also means a .pt file that works in one tool generally works in the others; it's the same underlying model, just wired into a different node interface.

Inputs and output

Just two required inputs - image and model - which is refreshingly simple next to BMAB's bigger sampler-driven detailer nodes. The output is a single MASK, which you wire into anything that takes a mask: a manual KSampler set up for masked img2img, ImageCompositeMasked, or straight into BMAB's own resize/fill nodes if they happen to accept one.

Installing it

ComfyUI Manager: search comfyui_bmab, install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/portu-sim/comfyui_bmab.git
cd comfyui_bmab
pip install -r requirements.txt

plus the two companion packs the README lists as required - comfyui_controlnet_aux (Fannovel16) and ComfyUI_IPAdapter_plus (cubiq), installed the same clone-then-pip way.

Where it gets used and where it doesn't

The README doesn't document exactly where these .pt files need to live on disk, and the dropdown simply reflects whatever BMAB finds. If it comes up empty, the fastest fix is usually installing Impact Pack or ADetailer's model set through ComfyUI Manager first - since they share the same weight family, whichever folder ComfyUI Manager drops them into is very likely the one BMAB is already scanning too - rather than hunting down each .pt file by hand.

The other thing worth knowing: this node just detects. If you're expecting it to also fix whatever it finds, you want one of BMAB's Face Detailer, Person Detailer, or Hand Detailer nodes instead - those wrap this exact detection step in a full crop-redraw-composite pipeline. Use BMAB Detector when you specifically want the mask and nothing else, or when you're building a custom pipeline BMAB's own detailers don't cover.