FIND EDGES (JOV) 🔳

FIND EDGES (JOV) 🔳 does one thing: pulls the outlines out of an image and gives them to you twice - once as a picture you can look at, once as a 😷 MASK you can actually use. That mask output is the whole point. Edge masks feed inpainting, controlnet-style conditioning, sketch-to-image workflows, and composite masking; a lot of ComfyUI edge tools make you extract a mask as a second step, and this one hands it over directly.

Quick honesty before you get attached to the name: this node is the pre-2.0 version. In Jovimetrix 2.x the edge-detection functionality was folded into a reworked node called ADJUST: EDGE (JOV) as part of the big v2.0/2.1 reorganization (the README's changelog documents the split of the old mega-ADJUST node into BLUR / EDGE / LIGHT / PIXEL). The old FIND EDGES name may still show up in saved workflows, but the maintained path today is ADJUST: EDGE - same job, plus an algorithm picker (Canny and friends), radius, and iteration controls. If you're starting fresh, reach for that one.

How it works

The mechanism is classic edge detection: it evaluates local contrast and marks where intensity changes sharply, then thresholds those responses so you get clean outlines instead of noise. The two outputs mean you don't have to choose between visualizing and using:

• 🖼️ IMAGE - the edges rendered as a viewable image. Good for eyeballing whether your threshold is sane.
• 😷 MASK - the same edges as a mask, ready to plug into anything that accepts masks.

The settings are sparse on purpose: LoHi is a two-value vector (low/high threshold, 0–1) that controls how sensitive detection is - low threshold catches faint edges, high threshold keeps only the strong ones - and the other float (🔳) is the strength/response dial. If the edges come out too busy, raise the threshold; if they vanish, lower it.

Installing it

It ships in Jovimetrix, so the install is the pack install:

# ComfyUI Manager: search "Jovimetrix" and install
cd ComfyUI/custom_nodes
git clone https://github.com/Amorano/Jovimetrix.git
cd Jovimetrix
pip install -r requirements.txt

Restart ComfyUI. It pulls in opencv-contrib-python - that's where the edge algorithms actually run - along with the pack's other dependencies. No model downloads.

Troubleshooting

• Everything is noise - your low threshold is too low. Raise LoHi's low value.
• Almost nothing detected - thresholds too high, or the image is already soft. Lower them, or blur first.
• You see the node but it won't load - this is the legacy name. If it errors on load after a pack update, you've hit the 1.7.48→2.x split; rewire onto ADJUST: EDGE (JOV) and delete the stale node.

That last point is the real gotcha and worth repeating: Jovimetrix 2.0 removed a bunch of node names, and the README is explicit that you shouldn't update past 1.7.48 if you depend on the old ones. Edge detection, though, survived the cut - it just got a new name and more knobs.