UltralyticsDetector (Pipe)(ultralyticsDetectorForDetailerFix)

Description

The UltralyticsDetector (Pipe) node creates a bounding-box detector from an Ultralytics YOLO model and packages it as a PIPE_LINE for use with the Easy-Use DetailerFix pipeline. It belongs to the EasyUse/Fix category and is designed for workflows where a detected object is repaired or refined by a later fix stage. The node does not perform the repair itself; it only supplies the detector configuration as part of the pipeline pipe.

Inputs

model_name

• Type: COMBO
• Required: True

Selects the Ultralytics detection model to load. The dropdown lists the Ultralytics models available to ComfyUI. Choose the model that should be used to locate objects in the image.

bbox_threshold

• Type: FLOAT
• Required: True
• Default: 0.5
• Min: 0
• Max: 1
• Step: 0.01

The confidence threshold for detected bounding boxes. Only detections with confidence above this value are kept. Raise the threshold to reduce false positives; lower it to include more potential detections.

bbox_dilation

• Type: INT
• Required: True
• Default: 10
• Min: -512
• Max: 512
• Step: 1

Expands each detected bounding box by this many pixels before it is used by the detailer fix. Positive values enlarge the box around the detected object, adding useful context. Negative values shrink the box, which can help when the detector returns boxes that are too loose.

bbox_crop_factor

• Type: FLOAT
• Required: True
• Default: 3.0
• Min: 1.0
• Max: 10.0
• Step: 0.1

Controls how much the detected bounding box is scaled when the crop is prepared for the fix stage. Larger values provide more surrounding context; smaller values keep the crop closer to the original bounding box.

Outputs

• PIPE_LINE: A pipeline object containing the configured Ultralytics bounding-box detector and its settings. Connect this to the pipe input of a downstream DetailerFix / repair node.

Usage Notes

• This is a detector provider, not an image processor. It returns no image or mask; it places the detector configuration into a pipeline pipe.
• Start with bbox_threshold around 0.5 and adjust based on missed detections or false positives.
• Use bbox_dilation to ensure the bounding box contains enough of the object for the fix stage. Negative dilation is available if detected boxes are too large.
• bbox_crop_factor controls how much surrounding context is included in the fixed region. Larger values include more context but may also cause the fix stage to process a much larger area.
• In a typical Easy-Use DetailerFix graph, feed the PIPE_LINE from this node into a fix node, then supply the image to be processed at the fix stage.