Remove Background

Strips the background out of an image and hands you back a cutout plus a mask. Compared to the swap and restore nodes in this pack, this one's refreshingly plain - there's no model dropdown, no dependency on the face-detection stack, just one input and it runs.

What's happening under the hood

Background removal as a category has three flavors people conflate constantly: post-generation removal (run a segmentation model over a finished image and guess where the edges are), native transparent generation (bake alpha into the image while it's being created), and targeted masking (select one specific object rather than "the foreground"). This node is the first kind - a single segmentation pass, no configuration exposed. The pack's README lists a lone rmbg.pth in its default model directory tree, and that's the model doing the work here; it's whatever FaceFusion itself bundles for this job, not a name you'll find benchmarked against the community's usual suspects.

Which matters if raw cutout quality is your actual goal rather than a convenience inside a face-swap pipeline. As of mid-2026 the community's default recommendation for background removal is BiRefNet - it shipped natively into ComfyUI's core in May 2026 - with InSPyReNet still winning some head-to-head comparisons on hair and fur specifically. Neither of those is what this node runs. If you're pulling this pack in purely for background removal and edge quality on flyaway hair or semi-transparent material actually matters to you, it's worth comparing this node's output against a dedicated BiRefNet or InSPyReNet node on your hardest image before standardizing on it. If you're already deep in a Faceless pipeline and just want a clean cutout without adding a second custom-node pack, this one's convenient exactly because there's nothing to configure.

Inputs and outputs

• images - the input image(s). That's the whole input list.

Two outputs:

• IMAGE - the cutout.
• MASK - the segmentation mask on its own, useful to route into an inpaint node, a compositing step, or anywhere else that wants a mask without you having to derive one separately.

Installing it

Manager: search Faceless Node for ComfyUI. By hand:

cd ComfyUI/custom_nodes
git clone https://github.com/jeffy5/comfyui-faceless-node
pip install -r requirements.txt

The model (rmbg.pth) goes in ComfyUI/models/faceless/, same as every other model this pack uses - not ComfyUI's regular folders. Run the bundled download_models.py (plain, or --all for every model the pack ships) and restart.

Where people get burned

Nothing to tune. Because there's no model choice or setting exposed, when the cutout struggles on hair or a semi-transparent object, there's genuinely nothing to adjust from this node's inputs - that's a limit of the single bundled model, not a setting you're missing.

Model not downloaded. The most likely failure is simply that rmbg.pth isn't in place yet - run download_models.py and restart before assuming anything else is wrong.