Nodes/ComfyUI-Grounding/Grounding Detector
ComfyUI Node

Grounding Detector

A ComfyUI node in grounding with 14 inputs and 4 outputs.

By PozzettiAndrea·Created 10 months ago·Updated 23 days ago· 45
Grounding Detector
  • model
  • image
  • bboxes
  • annotated_image
  • labels
  • masks
promptperson . car . dog .
confidence_threshold0.30
single_box_modefalse
single_box_per_prompt_modefalse
bbox_output_formatlist_only
seed42
text_threshold0.25
florence2_max_tokens1024
florence2_num_beams3
yolo_iou0.45
yolo_agnostic_nmsfalse
yolo_max_det300
Categorygrounding

Inputs (14)

NameTypeDefaultDescription
modelGROUNDING_MODEL
imageIMAGE
promptSTRINGperson . car . dog .Period-separated (.) = multiple objects: 'banana. orange' finds bananas AND oranges. Comma or no separator = single object: 'banana, orange' finds items labeled 'banana, orange'
confidence_thresholdFLOAT0.300–1Confidence threshold for detections. Typical: 0.2-0.35 (permissive), 0.35-0.5 (balanced), 0.5+ (strict)
single_box_modeoptBOOLEANfalseReturn only the highest-scoring detection. Use for referring expressions (e.g., 'the red car on the left')
single_box_per_prompt_modeoptBOOLEANfalseReturn highest-scoring detection for each prompt/label (e.g., 'banana. orange' returns best banana and best orange). Ignored if single_box_mode is True
bbox_output_formatoptCOMBOlist_onlylist_only: SAM2-compatible | dict_with_data: includes labels/scores
seedoptINT420–4294967295Fixed seed for reproducible results (affects mask visualization colors and model randomness)
text_thresholdoptFLOAT0.250–1GroundingDINO ONLY! Text confidence threshold
florence2_max_tokensoptINT10241–4096🌸 Florence-2 ONLY! Max tokens for generation
florence2_num_beamsoptINT31–10🌸 Florence-2 ONLY! Beam search width
yolo_iouoptFLOAT0.450–1🌍 YOLO-World ONLY! IoU threshold for NMS. Typical: 0.3-0.4 (keep more overlapping boxes), 0.45 (balanced/default), 0.5-0.7 (aggressive filtering)
yolo_agnostic_nmsoptBOOLEANfalse🌍 YOLO-World ONLY! Class-agnostic NMS. Enable when detecting overlapping objects of different classes (e.g., person holding bottle)
yolo_max_detoptINT3001–1000🌍 YOLO-World ONLY! Max detections per image. Typical: 100 (sparse), 300 (balanced/default), 500-1000 (dense/crowded scenes)

Outputs (4)

NameTypeDescription
bboxesBBOX
annotated_imageIMAGE
labelsSTRING
masksMASK