Load Face Model 🌌 ReActor

Once you've been face-swapping for a while, feeding the same source photo into every workflow gets old - and every fresh photo means ReActor re-analyzes the face from scratch. ReActor Load Face Model (class ReActorLoadFaceModel) is the shortcut. It loads a face model - a tiny saved file that already contains the extracted identity - from a dropdown, and hands it downstream as a FACE_MODEL. No source image, no re-detection, no hunting for that one good reference shot. Just pick the face you saved and go.

A face model here is a .safetensors file holding the InsightFace identity embedding (the ArcFace vector - facial geometry, not a picture). You created it earlier with Save Face Model or Build Blended Face Model, and it lives in ComfyUI/models/reactor/faces. These files are lightweight, portable, and - the underrated part - reproducible: the same model gives you the same identity input every run, which a photo-plus-detection path can subtly not do.

How it works

There's no image processing in this node. It reads the file you select and deserializes the stored embedding into a FACE_MODEL object. That object plugs into the face_model input of a main ReActor swap node, where it replaces the source_image - instead of "analyze this photo for a face," you're saying "use this exact identity I already computed." It can also feed Make Face Model Batch, which is how you stack several saved faces together to blend them.

The inputs and outputs that matter

This is about as simple as a node gets:

• face_model - a dropdown of every .safetensors face model in ComfyUI/models/reactor/faces. If you've never saved one, the only entry is none, and that's the tell that you need to run Save Face Model first.

The single output is FACE_MODEL. It wires into either the face_model input of a swap node (to swap using that identity) or into a Make Face Model Batch node (to combine it with others). That's the whole surface.

How to install it

It's bundled in the ReActor pack, so installing ReActor gives you this node. ComfyUI Manager: Install Custom Nodes, search ReActor, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/Gourieff/comfyui-reactor-node

then run install.bat / install.py and restart. This node itself downloads nothing - face models are files you generate, not something you fetch. The heavy dependencies (InsightFace, the swap and restoration models) belong to the swap nodes this one feeds.

Common issues & troubleshooting

The dropdown only shows none. You haven't saved any face models yet, or they're in the wrong folder. Run Save Face Model (feed it an image or a face model, give it a name) and confirm the .safetensors file lands in ComfyUI/models/reactor/faces.

I saved a model but it's not in the list. ComfyUI only scans that folder when the page loads. Refresh your browser and the new model appears. The README's own aside is worth heeding here: use ComfyUI Manager so a refresh doesn't wipe an unsaved workflow out from under you.

The loaded face doesn't look as accurate as I expected. A face model is only as good as what built it. A model made from one so-so photo carries that in; if likeness matters, build a blended model from several clean, well-lit shots of the person via Build Blended Face Model, then load that instead.

The node won't load / the ReActor menu is missing. That's the pack's InsightFace dependency failing to build, not this node - historically it needed Visual Studio C++ Build Tools on Windows and is the number-one thing that breaks this pack. Fix it at the pack level with the prebuilt Insightface wheel the README links for your Python version. And note the standing caveat: everything the ReActor chain does rides on InsightFace's non-commercial weights, so it's personal and research use, not something to ship in a paid product.