New node: Resize To Canvas Size (COI)
- Computes a Centre of Interest from an optional mask (bounding box centre
or weighted centroid/centre of mass) and places it at a user-selected
point on the canvas
- 7x7 target grid (49 dots overlaid on the 3x3 nonant background) lets the
user pick any canvas fraction in steps of 1/6; default (3,3) = canvas centre
- Three anchor modes: manual (image centre), mask_bbox_centre, mask_weighted_centre
- Graceful fallback to image centre when no mask is connected or mask is all zeros
- Mask is auto-resized if its dimensions differ from the source image
- All existing fill/scale/padding options carried over from the original node
Also adds:
- examples/ResizeToCanvasSize_Example_Workflow.json demonstrating both nodes
side by side with a face-detection COI pipeline (UltralyticsDetectorProvider
-> BboxDetectorSEGS -> SAMDetectorCombined -> ResizeToCanvasSizeMask)
- Updated README documenting both nodes, the 7x7 grid concept, COI modes,
and a face-centred portrait crop recipe
Carasibana
92899ca312