🧭 Mesh UV Watertight

What it's actually for

Every 3D pipeline needs one node that answers a single question: is this thing actually closed, and does it have usable UVs? Before you remesh, before you print, before you weld - you want the numbers, not a guess based on how the viewport looks.

This is that node. Be clear about what it is, though: in the shipped version it's a diagnostic pass-through. The mesh goes in, the same mesh comes out, and the useful output is the print in your console. That's not a criticism so much as a calibration - the node's whole value is the console readout.

How it works

The watertight check is stricter than trimesh's default. It requires both mesh.is_watertight and a Euler number of 2 - the topology signature of a single closed surface with no handles. So a mesh that's closed but has an extra tunnel through it (a hole-drilled solid, or a generated shape that grew a loop) reads as not watertight. If that check throws, it falls back to plain is_watertight, and if that throws too, it reports false.

The UV side extracts mesh.visual.uv, sanitises NaNs, and reports the min/max bounds, whether the layout sits inside the unit square, and an island count. That last number is measured with a rough heuristic rather than real island analysis - it walks the UV array, finds jumps in UV distance larger than mean+σ, and counts them. It's a smell test, not a census. A layout that's been sorted or re-indexed will fool it.

Point clouds and empty meshes are short-circuited: if there are no faces, the node returns immediately without pretending to have an opinion.

The show_checker, show_wireframe, show_seams and show_chart_ids options exist in the node's schema and are computed into the UI payload the code builds - but the shipped preview() returns the mesh alone, so in practice those toggles don't reach a viewer. If you were expecting a checker-textured viewport, use preview_mesh_uv or Ultimate3DViewSaveLoad instead.

The inputs that matter

• mesh - the TRIMESH to inspect. That's the only required one.
• show_seams / show_wireframe / show_chart_ids - present in the schema; currently inert in the shipped code. Set them if you like; watch the console for the report either way.

One output: mesh, unchanged. Slot it inline in a chain - inspect, print, pass through - by wiring the output to the next node. Because it's an output node it always runs, so you get a fresh readout on every queue, which is genuinely handy at the end of a repair chain.

The console output is the deliverable:

[Mesh] V=48213 F=96422 watertight=False
[mesh_uv_watertight] Preparing preview

Install

cd ComfyUI/custom_nodes
git clone https://github.com/Antonioilev/ComfyUI_Antonioilev_Lightpack.git
# restart ComfyUI

Manager search: "Antonioilev Light Pack". No requirements.txt ships with the repo and __init__.py swallows import errors behind a loaded/failed count, so a missing node means a missing Python package. This one needs trimesh, plus folder_paths (ComfyUI itself) for the output directory - no extra pip work in a normal install.

Where people get burned

• Unwelded meshes always fail the watertight test. If your generated asset reports watertight=False with a hole count in the hundreds, that's chart splits, not holes. Weld first (AntonioilevUVWatertight), then re-run this and watch the verdict flip.
• The Euler-number check is strict. A mesh you know is solid can report false if it has an internal loop. If that surprises you, that's the node doing its job.
• The island count is approximate. Treat it as "how chaotic is this UV array" rather than a real count; for a genuine number, run xatlas and count what comes back.
• Nothing appears in the UI. Expected. This node's interface is stdout - keep the ComfyUI console visible, or you'll see a node that seems to do nothing at all.