K2 Region from BBox

K2 Region from BBox turns a detector's bounding box into a K2 region. Where K2 Region asks you to type pixel coordinates, this node reads them from a detector - YOLO, the Krea 2 bbox prompting from KJNodes' Ideogram-style prompt builder, or this pack's own person selectors - and hands you a ready region with a name, a prompt, and a role.

This is the automation bridge for the K2 Lab suite in comfyui-FVMtools. If you're doing character-consistent multi-person work, you don't want to hand-draw the same boxes every run; you want a detector to find the faces and bodies, and this node to convert those detections into the regions the attention router understands.

The inputs that matter

• name and prompt - same as on K2 Region: the label used in spatial instructions and reports, and what appears inside the box.
• bboxes (KJNodes-style BBOX list) or bounding_box (standard BOUNDING_BOX) - the detection input. bounding_box takes precedence when both are connected.
• bbox_index - which box in the list to use, zero-based. Out of range clamps to the last entry.
• bbox_format - xyxy (left/top/right/bottom) is what YOLO, BOUNDING_BOX, and the KJ prompt builder emit; xywh is left/top/width/height. auto tries xyxy first and only falls back to xywh when xyxy yields an impossible box - but it can't always tell them apart, so set it explicitly when you know the format.
• reference_width / reference_height - the canvas the normalized (0..1) coordinates are expanded against. Must match the render size.
• grow_px - expand (positive) or shrink (negative) every side of the box. Handy when a detector's tight face box needs to become a head or body region.
• role and priority - same meaning as on K2 Region: subject/background/auto, and compile/face-claim order (not strength).

Chain through regions like any other region node; identity_prompt is available for face descriptions.

Typical wiring

Detector → BBOX → K2 Region from BBox → K2 Compose. Run it, look at the result with K2 Region Preview, and adjust grow_px until the box actually covers what you meant. Detector boxes are usually tighter than you want for a subject region - a face detector gives you a face box, and your prompt probably describes the whole person, so grow_px is where you'll spend your time.

Install

Same as the rest of the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/ping1979ping/comfyui-FVMtools

Restart ComfyUI, find it under FVM Tools/K2. No extra Python dependencies; the detector itself is whatever your workflow already uses.