Load Face Model

This is the "give me that person again" node. If you've saved face models with Build Blended Face Model or Save Face Model, they live in ComfyUI/models/simswap/faces/ as .safetensors files, and this node is the dropdown that loads them. Pick a name, out comes a FACE_MODEL, wire it into the Simswap swap node's face_model input, done. No source photo needed in the workflow at all.

It's a deliberately thin node, and that's the point. The alternative is dragging a reference image into every single workflow that wants to swap in the same person. Save once, and every future graph just needs a Load Face Model with the right name. That's also the payoff of the whole save/load round-trip in this pack: the identity becomes a file, not a dependency.

What it does

One input, one output:

• face_model - a dropdown listing none plus every .safetensors file found in ComfyUI/models/simswap/faces/. none is the default and the "I don't have a saved model" escape - pick it and the node passes out an empty model, which is equivalent to not wiring a face model at all.
• FACE_MODEL output - feed it to the swap node's face_model slot (it's a drop-in alternative to source_image), or into Save Face Model if you're re-exporting it.

Mechanically it's simple: safe_open reads the saved tensors back off disk and rebuilds an InsightFace Face object - bounding box, keypoints, pose, and the ArcFace embedding that carries the identity. The swap node then treats it exactly like a freshly-detected face, so behavior is identical to passing a source_image, just cheaper and more stable.

Installing it

Standard for this pack:

cd ComfyUI/custom_nodes
git clone https://github.com/TaiTair/comfyui-simswap
# or: ComfyUI Manager → search "simswap" → install
# then restart ComfyUI

No extra models to fetch for this node itself - it reads files other Simswap nodes (or you, manually) dropped in models/simswap/faces. First start does install the pack's pinned dependencies (insightface==0.7.3, onnxruntime, onnx, opencv-python, numpy, segment_anything), and the swap pipeline will want a detection model under models/simswap before anything works.

Gotchas

• If the dropdown is empty except for none, nothing's been saved yet - run Build Blended Face Model (or Save Face Model) once to create a file.
• Deleting or renaming a .safetensors in that folder while ComfyUI is running won't refresh the dropdown until restart. Cheap gotcha, but it bites.
• As with the rest of the pack: it's a personal, unmaintained port, so don't expect updates to the format. A saved model is a .safetensors with insightface's Face tensors inside - nothing exotic, just don't hand-edit them.