UV Seams From Islands

UV Seams From Islands is a neat trick for anyone who has ever stared at an unwrap and thought "if only the cuts were where the islands already are." It looks at your existing UV islands - the separated chunks - and marks their boundary edges as seams automatically. Instead of hunting edge loops by hand to define cuts, you're saying: the layout I already have is the layout I want, make it permanent.

How it works

It's the auto-generated wrapper for bpy.ops.uv.seams_from_islands on your BPY_OBJ: edit mode, run the operator, return the same object. Two booleans:

• mark_seams (default on) - mark the island boundary edges as seams.
• mark_sharp (default off) - additionally mark those edges as sharp, which matters if you're doing any normal-based shading or beveling later.

That's the whole node. Output is the BPY_OBJ you fed in.

Where it fits

The real-world use case is converting a good unwrap into a re-usable seam layout. Say you've got an avatar layer where the islands are placed well but you want to re-unwrap with a different method - the seams from your current islands are the natural cut pattern, so run Seams From Islands first, then unwrap, and you'll get something close to your current layout but with freshly-computed UVs. It's also the fastest way to set seams for a mesh that came in with no UVs and no obvious topology landmarks: do a quick projection to break it into islands, run this, and you have instant seams for a proper unwrap.

It pairs naturally with UV Mark Seam - where that node wants you to pick edges, this one derives them. If you're hand-authoring the cut layout you'll still use Mark Seam; Seams From Islands is for when the layout should drive the seams.

Installing Avatar Graph

This node ships in the Avatar Graph pack from Avatech. From ComfyUI/custom_nodes:

git clone https://github.com/avatechai/avatar-graph-comfyui
cd avatar-graph-comfyui
python -m pip install -r requirements.txt

then restart ComfyUI with --enable-cors-header. Or ComfyUI Manager → search Avatar Graph → Install.

The usual traps: the pack pins bpy==3.6.0, which needs Python 3.10.x (anything else fails the pip install and gives you no nodes). First launch auto-downloads the ~2.5 GB SAM model. And the pack is openly still in flux - pin it once your workflow works.