TBG SAM3 Segmentation

A ComfyUI node in TBG/SAM3 with 13 inputs and 7 outputs.

By Ltamann·Created 9 months ago·Updated 17 days ago· 167

◄confidence_threshold0.40►

◄pipeline_modeall►

◄detect_alltrue►

◄max_detections50►

◄instancesfalse►

◄crop_factor1.5►

◄min_size100►

◄fill_holesfalse►

◄text_prompt►

CategoryTBG/SAM3

Inputs (13)

Name	Type	Default	Description
sam3_model	SAM3_MODEL		SAM3 model loaded from LoadSAM3Model node
image	IMAGE		Input image to perform segmentation on
confidence_threshold	FLOAT	0.400–1	Minimum confidence score to keep detections. Lower threshold (0.2) works better with SAM3's presence scoring
pipeline_mode	COMBO	all	Which prompts from pipeline to use.
detect_all	BOOLEAN	true	When enabled, detects all objects. When disabled, uses max_detections value.
max_detections	INT	501–100	Maximum detections when detect_all is disabled.
instances	BOOLEAN	false	When ON: keep only detections whose boxes overlap a positive box or contain a positive point. When OFF: return all SAM3 detections including instances.
crop_factor	FLOAT	1.51–4	Crop factor used when building combined SEGS (Impact Pack style). 1.0 = tight bbox.
min_size	INT	1001–500	Minimum segment size in pixels as a square side. 1=1x1, 200=200x200; smaller masks are discarded.
fill_holes	BOOLEAN	false	When enabled, fills holes inside each mask (solid segments).
text_promptopt	STRING		Text to guide segmentation (optional).
sam3_selectors_pipeopt	SAM3_PROMPT_PIPELINE		Unified pipeline containing boxes/points).
mask_promptopt	MASK		Optional mask to refine the segmentation.