What's this?

I came across this super cool object detection dithering effect posted on Reddit (using TouchDesigner): Object detection dithering effect

Inspired by that, I tried recreating a similar effect in ComfyUI.

While it doesn’t come close to TouchDesigner in terms of performance or flexibility, I hope this serves as a fun demonstration of what’s possible in ComfyUI! ✨

Workflow

• 🟥 Webcam Capture node grabs visuals from Google Street View.
• 🟩 YOLO-World detects specific objects — in this case, human and car — and outputs bounding boxes.
• 🟪 Pixelation Process:
• Background is reduced to a limited color palette.
• Detected objects are rendered in grayscale using a mask.
• 🟨 Bounding Boxes and Labels as Mask:
• The image output by the YOLO node includes the background, bounding boxes, and labels all together, which means you can’t isolate just the boxes and labels directly.
• To work around this:
• Use Image Blend with the "difference" mode to subtract the input image from the YOLO output. Matching areas turn black; differences remain visible. As a result, the only visible elements are the bounding boxes and labels.
• Convert this result into a mask using ImageColorToMask.
• This technique unfortunately requires running two identical YOLO nodes — one for processing, and one just to extract the overlay — which is admittedly wasteful but necessary for now.