Bounding Box Visualizer

This is the honest little utility in the ComfyUI-Grounding pack: it draws the bounding boxes from a detector back onto your image and returns the annotated result. One input, one output, one optional number. There's no model involved, nothing downloads, no VRAM.

To be blunt, GroundingDetector already gives you an annotated_image preview, so if that's all you need you're done before you open this node. BboxVisualizer earns its keep in two situations: you want a different line width than the detector's built-in rendering, or you've taken the bboxes apart (filtered, reordered, hand-edited) and need to redraw the results on the original frame. It handles both output formats the detector can produce - list_only and dict_with_data (where it can also stamp labels and scores onto the boxes).

The inputs

• image - the frame to draw on.
• bboxes - the BBOX wire from GroundingDetector (either format works).
• line_width - 1 to 10, default 3. The only knob, and it's self-explanatory.

Output is annotated_image, which you'd route to a preview or save node to eyeball what the detector found before you commit to segmenting or cropping on those boxes.

Installing it

It's part of the same pack, so the install is the shared one:

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

Restart ComfyUI (or search "Grounding" in Manager) and the node appears under the grounding category. No separate models or downloads - it's pure PIL drawing.

Where people get burned

There's not much to burn here. The one thing to know: the node draws red boxes with the width you set, and if you feed it a list_only bbox batch it has no labels to show. If you want labels on the boxes, run the detector in dict_with_data mode. Otherwise this is as close to zero-friction as a ComfyUI node gets - a few seconds of setup and you've got a visual check on anything you're about to cut, crop, or mask.