YOLOv8n Person Seg

When you need a person cut out of an image - not just a box around them, but an actual silhouette - YOLOv8n Person Seg is the node. It's the nano Ultralytics person-segmentation model running on Runware's cloud, and it returns the detected people as a mask IMAGE with edge controls. The "nano" part means it's the fast, cheap tier: good enough for compositing and mask-based editing, not the highest-fidelity segmentation on earth.

The use case triangle it serves is the one the ecosystem keeps circling: background replacement (person mask → new background), selective inpainting (mask the person, regenerate everything else), and subject isolation for further processing. Locally, person segmentation means installing a segmentation model plus its plumbing; here it's one node with one image input and a mask out. If you just need "give me the people as masks," this is the shortest path in the pack.

How it works

An imageMasking task on Runware's cloud (runware:35@4): upload the image via the SDK, request over REST, and the resulting segmentation mask downloads into an IMAGE tensor. The settings shape the mask quality:

• settings.confidence (0.5) - detection threshold for people.
• settings.maskBlur (5) - edge-smoothing radius; a little blur hides the cut-out seam when you composite.
• settings.maskPadding (10) - extend or shrink the mask by pixels. For cutting out a person you often want slight positive padding so no hair or clothing edge gets clipped.
• settings.maxDetections (6) - cap on people masked, highest confidence first.

Output is image (IMAGE) - feed it into compositing, background-removal, or mask-based inpaint workflows.

The inputs that matter

Required input: image. For a clean cut-out: confidence 0.5, maskPadding +10, maskBlur 5, then composite. Crowd shots hit the maxDetections cap; raise it if you're segmenting a group.

Install and API key

Install once for the whole pack:

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

Restart ComfyUI (or install Runware from ComfyUI Manager). API key from runware.ai/api-keys, set in ComfyUI Settings → Runware API key, exported as RUNWARE_API_KEY, or via runware auth login.

Gotchas

Nano-tier segmentation has known weak spots: flyaway hair, thin clothing edges, and people overlapping each other. For those cases the honest advice is that dedicated background-removal models (BiRefNet and friends) are usually better at hair than any YOLO segmentation, so if your goal is a clean person cut-out with hair intact, that's a different node than this one - this node is for "person mask, fast and cheap." Overlapping people also merge into one blob; raise confidence or separate subjects. And the cost reality: YOLO is open source and free to run locally, so on a big batch the per-image cloud price will eventually make you install it locally - this node is for the occasional mask or when you're already paying for the cloud stack.