Quickstart

Five minutes from install to your first result.

1. Install

ComfyUI Manager (recommended)Manual (git)

1. Open Manager → Custom Nodes Manager.
2. Search for ComfyUI-JH-PixelPro and click Install.
3. Restart ComfyUI.

cd ComfyUI/custom_nodes
git clone https://github.com/jetthuangai/ComfyUI-JH-PixelPro.git
cd ComfyUI-JH-PixelPro
pip install -r requirements.txt

Use the same Python that runs ComfyUI (for a portable or venv install, its own python). Then restart ComfyUI.

Requirements

Python 3.10+, the PyTorch that comes with ComfyUI, Kornia ≥ 0.7, MediaPipe ≥ 0.10, OpenCV ≥ 4.8 and SciPy ≥ 1.10. Optional: onnxruntime-gpu (or onnxruntime) for Depth Anything and Depth Lens Blur. An NVIDIA GPU with 8 GB VRAM or more is recommended. Everything also runs on CPU, just slower.

2. Find the nodes

Right-click the canvas (or double-click and search), then open Add Node → ComfyUI-JH-PixelPro. Nodes are grouped the same way as this site:

color · mask · face · compositing · filters · geometry · looks · qa

Tip

Type part of a node name in the search box, for example Frequency, Matte or CLAHE.

3. Run your first workflow

1. Copy workflows/sample_portrait.jpg from the pack folder into ComfyUI/input/.
2. Drag a sample workflow onto the canvas. Good first picks:
   • CLAHE Local Contrast example: one node, instant result.
   • Face Detect Lite example: face box, landmarks and alignment.
   • Mask Cleanup example: also copy workflows/sample_binary_mask.png into ComfyUI/input/.
3. Press Queue Prompt. Change one setting at a time and compare.

All sample workflows are listed in Examples.

4. Build your own chain

A good way to learn the pack is one classic retouch step per group:

Goal	Try
Even out skin but keep pores	Frequency Separation or Skin Smoother
Sharpen eyes and hair only	Detail Mask → Smart Sharpen
Grade colour like Camera Raw	ColorLab, Tone Curve
Match one photo to another	Color Matcher or Color Matcher Region
Cut out hair cleanly	Trimap Builder → Alpha Matte
Run AI on a face, paste it back	Face Detect Lite → Facial Aligner → … → Unwrap Face
Check what changed	QA Image Diff

Troubleshooting

The nodes do not show up

Check the ComfyUI console for an import error from ComfyUI-JH-PixelPro. Usually a dependency is missing: run pip install -r requirements.txt with ComfyUI's own Python, then restart.

Face Detect Lite or Auto Align (LoFTR) downloads something

Yes, on first use only. Face Detect Lite downloads its YuNet weights (about 0.4 MB) and Auto Align with method = loftr downloads LoFTR weights (about 45 MB) into the torch cache. Face Detect (MediaPipe) downloads its model (about 5 MB) into ComfyUI/models/mediapipe.

The machine slows down on a big image

Turn on tile_mode in Skin Smoother for anything larger than a preview. Since v1.5.2 tiles are sized to keep memory near 2 GB. Close other heavy workflows if VRAM is already full.

I changed the pack but nothing changed in ComfyUI

ComfyUI loads custom-node code once at start-up. Restart ComfyUI after updating the pack.