🎯 Advanced Tracking Pro

The name promises a lot. AdvancedTrackingPro sounds like the multi-object tracking backbone of the whole kanibus/kanibus pack - face/body/object modes, re-identification, object counting. The reality, from the source (nodes/advanced_tracking_pro.py): it draws a hardcoded rectangle on your image and calls it tracking. No model runs, no objects are detected, no re-ID happens.

This is the second-most-dishonest node in the pack after EmotionAnalyzer (and in a tie with ObjectSegmentation and AIDepthControl, honestly). The repo is Claude-generated, last commit Aug 2025, and a bunch of its "AI" nodes are stubs wearing enterprise-grade names. This one's stub is egregious because the outputs look like they did work.

What it actually does

Look at track_objects: it converts your image, then ignores the image content entirely and picks from three hardcoded detection lists based on tracking_mode:

• face → one box labeled "face" at 0.9 confidence
• body → one box labeled "person" at 0.85
• objects/all → one face box plus one person box

It then draws those boxes onto the image, fills a rectangle mask, and returns object_count = how many hardcoded boxes there were. Same input, same output, every frame, no matter what's in the picture. The optional inputs previous_tracks, enable_reid, max_objects and confidence_threshold are accepted and largely ignored - confidence_threshold gets tweaked per "WAN version" (resolution) but never compared against any detection.

Outputs are: tracking_result (TRACKING_RESULT), annotated_image (IMAGE with boxes drawn), object_masks (MASK), object_count (INT).

The inputs that matter

Honestly: none of them. tracking_mode just selects which canned box set you get. The "count" will say 1 or 2 depending on the mode you picked. That's the entire surface area of this node.

Installing

cd ComfyUI/custom_nodes
git clone https://github.com/kanibus/kanibus
cd kanibus    # lowercase - README's "cd Kanibus" fails on Linux/Mac
pip install -r requirements.txt   # or requirements_minimal.txt if it clashes
python install.py

Restart ComfyUI, find it in the Kanibus category. It loads no models (that's the point), so the README's "5.6GB of required ControlNet models" doesn't apply.

What to do instead

If you need real multi-object detection and tracking in ComfyUI, use tools that actually run detectors - Ultralytics YOLO-based nodes (the pack even lists ultralytics as a dependency, but this node never touches it), or proper tracker packs. For the pack's own scope, the real tracking value is NeuralPupilTracker (eyes) and the MediaPipe pose/hand/face nodes. This node is the pack's version of a mock-up - useful only if you want a pretend detection pipeline to test wiring while you source a real one.

Troubleshooting

• Boxes don't match the people in your image - not a bug, that's the node. It's not looking at your image.
• object_count is always 1–2 - yes. It counts its hardcoded boxes.
• Node missing after install - check the console for a dependency import error (mediapipe, torch); fix and restart.

Bottom line: if a downloaded workflow contains this node, treat its outputs as placeholders and swap in a real detector before relying on anything downstream. It's a shell that dresses up a cv2.rectangle call as a tracking system.