Load Face Fix Detector (YOLO)
One detector, wired everywhere
- face_detector
The first thing you'll notice wiring up Ultimate Face Fix is that the fix node wants a face_detector connection, not a model picked inside a widget. That connection comes from this node - Load Face Fix Detector (YOLO) - which exists to load the YOLO face detector once and hand it out to every face-fix node that needs one.
Why a separate loader instead of a combo buried in the fix node? Because the same detector object is shared across the integrated Ultimate Face Fix node, the Extract node, and the Auto Denoise node. Load it once and they all reuse it instead of each spinning up its own copy - the same reuse pattern as CLIP and VAE loaders elsewhere in ComfyUI, just with a custom FACE_FIX_DETECTOR type on the wire. That's also why you can't plug it into an arbitrary IMAGE input: it's a typed handle, not a tensor, so it only connects to slots that expect a face detector.
Mechanically it's thin. A combo (detector_model) lists whatever YOLO files it finds in ComfyUI/models/face_fix/detectors/, and it wraps the one you pick in an Ultralytics-backed detector. The pack's setup script automatically downloads face_yolov9c.pt - a YOLOv9c face detector from Bingsu's ADetailer model line, the same family that standardized face detection across the A1111 and ComfyUI ecosystem. If the dropdown ever shows "none available," the model didn't land; rerun the prepare script or drop your own .pt/.safetensors YOLO into that folder and it'll appear.
Install is the shared pack setup: ComfyUI Manager → search Ultimate Face Fix → Install, then restart. Or manually:
cd ComfyUI/custom_nodes
git clone https://github.com/Merserk/ComfyUI-Ultimate-Face-Fix.git
cd ComfyUI-Ultimate-Face-Fix
pip install -r requirements.txt
python scripts/prepare_models.py --comfy-root ../..
One thing worth knowing before you build a product on this: the YOLO path runs on Ultralytics, which is AGPL-3.0, and the pack pins ultralytics==8.4.102 for good reason - a poisoned Ultralytics release reached ComfyUI users through a detailing pack in December 2024. Pinned is correct.
And yes, the fix node also accepts ComfyUI's built-in MediaPipe face model (Apache 2.0) as the optional face_landmarker - the license-clean route if AGPL gives you pause. But that's the landmarker slot, not a substitute for the YOLO face_detector. For the main detection step you still want this node, and it'll be the first node in your graph most of the time.
Inputs (1)
| Name | Type | Default | Description |
|---|---|---|---|
| detector_model | COMBO | 1 options: face_yolov9c.pt |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| face_detector | FACE_FIX_DETECTOR | — |