comfyui-gits-tracked-identity

ComfyUI GITS Tracked Identity

A ComfyUI custom-node package that places a tracked, camera-facing identity graphic over one or more faces — for single images, ordered video batches, and live webcam. The name describes the visual technique; this repository contains no copyrighted franchise artwork. Bundled cyan placeholders are original geometric art.

Version 0.2.0 — multi-face IDs, Temporal LaMa, webcam, glitch/signal effects, appearance matching, motion prediction, and improved temporal fill.

Everything runs locally. Dependencies and models use MIT/Apache-style licenses only (see Licenses).

Why the overlay is applied twice

video/image batch -> Advanced Tracking -> preprocessing / diffusion -> Final Artwork Overlay
                         |                                      ^
                         +-- exact RGB overlay + alpha mask ----+

GITS Face Tracking + Removal (Advanced) puts an opaque proxy plate plus the graphic into the images sent toward diffusion. It also returns the exact aligned RGB graphic and alpha separately. GITS Final Artwork Overlay applies those exact pixels after VAE decoding so letters, circles, thin lines, and ring rotation do not deform. Do not expect KSampler to reproduce logo art.

For a fast path with no diffusion, use Simple or Webcam: they track, optionally remove the face with LaMa, and composite the exact graphic in one node.

Artwork is optional on every node. If static_logo, static_logo_mask, ring_logo, and ring_logo_mask are all unconnected, tracking and LaMa face removal still run. The outputs are then the reconstructed background (face removed) with no graphic overlay.

Installation

Reference environment: Python 3.10 or 3.11.

Clone into ComfyUI

cd ComfyUI\custom_nodes
git clone https://github.com/Niutonian/ComfyUI-GITS-Tracked-Identity.git
cd ComfyUI-GITS-Tracked-Identity

Or copy this folder to ComfyUI/custom_nodes/ComfyUI-GITS-Tracked-Identity.

Dependencies

Use the same interpreter that starts ComfyUI:

# Embedded ComfyUI Easy Install (Windows example)
..\..\python_embeded\python.exe -m pip install -r requirements.txt

# Or a normal venv
python -m pip install -r requirements.txt

OpenCV and PyTorch are normally already provided by ComfyUI. Face removal also needs ComfyUI-RMBG (or an equivalent Big-LaMa provider) loaded in ComfyUI.

Models (explicit download only)

Nodes never download weights at queue time. Install models once, then restart ComfyUI. Search the node menu for GITS.

Where to place them

Create this folder if it does not exist:

ComfyUI/models/gits_tracking/

Expected files (exact names):

File	Purpose	Approx. size
face_landmarker.task	MediaPipe face landmarks (primary tracker)	~3.6 MB
face_detection_yunet_2023mar.onnx	OpenCV YuNet fallback (small / hard faces)	~0.2 MB
big-lama.pt	Big-LaMa face removal / inpaint	~196 MB

Full example on a typical Windows Easy-Install layout:

ComfyUI/
  models/
    gits_tracking/
      face_landmarker.task
      face_detection_yunet_2023mar.onnx
      big-lama.pt
  custom_nodes/
    ComfyUI-GITS-Tracked-Identity/

If you already have Big-LaMa from ComfyUI-RMBG at ComfyUI/models/RMBG/Lama/big-lama.pt, you can copy that file into gits_tracking/ instead of downloading it again. GITS prefers models/gits_tracking/big-lama.pt and can fall back to the RMBG path when configured through the LaMa remover node.

Manual download URLs

Download each file in a browser (or with curl / wget) and save it into ComfyUI/models/gits_tracking/ with the exact filename above:

1. Face Landmarker → save as face_landmarker.task
   https://storage.googleapis.com/mediapipe-models/face_landmarker/face_landmarker/float16/latest/face_landmarker.task

2. YuNet face detector → save as