Extract Edges (Canny)

How it fits

Typical backends

Caveats

• I/O contractThe output is a binary Mask, not a 3-channel Image, so a consumer expecting an RGB edge map (for example a diffusion ControlNet) needs a separate mask-to-image conversion.
• CompatibilityA non-greyscale input goes through a BGR-to-grey conversion before edge detection, so non-photographic inputs are still interpreted under the BGR colour assumption.
• FallbackAn empty Image short-circuits to an empty Mask without invoking the detector.
• Hard constraintThe Sobel aperture must be 3, 5, or 7; any other value aborts the call inside the detector.
• Parameter interactionThe low and high thresholds form a hysteresis pair: pixels above the upper threshold are always kept and pixels above the lower one only when connected to a strong edge, so setting them equal degenerates to a single threshold; the L2 gradient option trades a little speed for a more accurate gradient norm.
• State lifetimeAll parameters are captured once at first invocation; runtime changes have no effect and require a redeploy.