1.7 KiB
Intelligent AutoCrop
The production Intelligent AutoCrop node
(PortraitAutoCropPreview) removes uniform borders before subject framing.
It consumes an image and Input & Crop Config. The old
IntelligentAutoCrop and IntelligentAutoCropV2 IDs have been removed.
Removing solid black image-hosting borders, including Alamy-style borders, is an explicit purpose of this stage. The production node offers a conservative policy or literal configured tolerances and a shared safe crop rectangle for image batches.
Configuration
Settings are in the first, AUTOCROP · SOLID BORDER REMOVAL section of the configuration node:
autocrop_strip_bottom_banner: Detect a dark strip with bright content touching the bottom edge and remove it before border detection.autocrop_detect_borders: Enable uniform-border scanning on all four edges.autocrop_fuzz_tolerance: Allow color variation in a border.autocrop_edge_uniformity: Fraction of a line that must match its border color.autocrop_pad_px: Retain this many pixels of a detected border.
Smart Crop and resolution/aspect controls belong to the separate second section and are consumed by Smart Photo Prepare, not by this node.
Outputs
image, trim_left, trim_top, trim_right, trim_bottom, and detected.
Trim values are source-image pixels. This node does not output a transformed mask;
generate segmentation from the cropped image or crop existing masks consistently.
Detection analyzes a CPU copy where necessary; the output stays on the input tensor's device without image resampling. Border and banner detection remain heuristic, so inspect the result when tuning tolerances.