💜 Compositor Transforms Output V4

Compositor4TransformsOut is the V4 edition of the transforms decoder: it reads the compositor's layout JSON and emits x, y, width, height, angle, and the bbox set (bbox x, bbox y, bbox width, bbox height) for one layer as plain integers - the thing you feed into Set Area Conditioning, crop coordinates, or any node that wants a number instead of a drag gesture.

The one wrinkle is in the wiring. V3 exposed a dedicated transforms output on the Compositor; the V4 rewrite temporarily dropped that socket. What V4 does emit is fabricData_output - a STRING holding the canvas state JSON, which contains the same transforms and bboxes arrays in the same structure. So you wire Compositor4's fabricData_output into this node's transforms input. Same data, different door. If you're on V3, the V3 variant is the cleaner match.

Inputs

• transforms - the layout JSON (forceInput, so always wired). From Compositor4, feed it fabricData_output.
• channel - which layer to read, 1–8 (default 1). image1 is the background-most.
• forceInt - default on, returns everything as integers for conditioning nodes. Off gives you raw float precision.

Outputs

x, y (top-left, padding subtracted), width, height (scaled size), angle (rotation), and the axis-aligned bbox x/y/width/height for when the layer is rotated and you need its unrotated footprint. All INT.

Installing

Same pack, no models, no extra pip packages:

cd ComfyUI/custom_nodes
git clone https://github.com/erosDiffusion/ComfyUI-enricos-nodes.git

or ComfyUI Manager → "Compositor" → erosDiffusion. Restart to load.

Gotchas

• Numbers come from the last saved canvas state - drag something and the outputs won't move until the canvas saves (mouse out of the export area).
• Since V4 is still alpha, treat the "feed fabricData_output into transforms" wiring as the current workaround, not the permanent contract. When the author re-enables a dedicated transforms output on Compositor4, that's the cleaner port.
• angle is the rotation value; if a downstream node cares about the unrotated footprint, grab the bbox pair instead of width/height.