FF: Visualize Faces

This is the simplest node in the whole pack - two booleans, nothing else. That simplicity is the point: before you spend a run on an actual face swap, this draws exactly what FaceDetectorNode found and picked, so you can catch a bad detection or the wrong face selection before it costs you anything downstream.

It's worth telling this apart from its cousin FaceMaskVisualizer, because the names are close enough to confuse. This node is about the detection - did it find the right face, at the right box, with landmarks that look sane? FaceMaskVisualizer is about the mask an occluder or parser model would produce during a swap. Run this one first to confirm the detector got the right person; run that one once you're tuning mask settings for the swap itself.

How it works

It reads a FACE_DATA object - the same one FaceDetectorNode produces - and draws the detection results directly onto the image: a bounding box around the detected face, and the landmark points (eyes, nose, mouth corners, and similar reference points) the detector located. Nothing computationally interesting happens here; it's pure visualization, which is exactly what makes it cheap to drop into a graph as a sanity check.

Inputs and outputs that matter

• face_data (FACE_DATA) - from FaceDetectorNode.
• draw_landmarks (BOOLEAN, default true) - toggle the landmark points on or off.
• draw_bbox (BOOLEAN, default true) - toggle the bounding box on or off.

Output is a single IMAGE. There's no swap, no mask, no model beyond what detection already ran - this node is as close to free as anything in the pack gets.

Installing it

ComfyUI Manager: search Facefusion_comfyui, install, restart. Manually:

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

Full restart required. This node doesn't need any model downloads of its own - it only draws on top of whatever FaceDetectorNode already found, so as long as detection is working, this one will just work.

Common issues

Landmarks look scattered or clearly off the face. That's almost always a low-confidence detection sitting right at your score_threshold cutoff back in FaceDetectorNode. Raise the threshold there, or check the source image for the usual culprits - poor lighting, a small or heavily angled face, partial occlusion.

The box is right but nothing else changes when you toggle the booleans. Make sure you're actually re-running the graph after flipping draw_landmarks or draw_bbox - these are plain widget toggles with no other logic attached, so a stale cached output is the likely explanation if nothing visibly changes.

Nodes missing after install. This is a pack-wide issue, not specific to this node: confirm pip install -r requirements.txt completed inside ComfyUI's own Python (portable installs need python_embeded's pip), then fully restart rather than refreshing the browser.