Decimate Fast Simplification (backend)

If CGAL Edge Collapse is the decimation backend that obsesses over quality, Fast Simplification is the one that gets out of your way. It's the quick-and-dirty backend in GeometryPack's Decimate Mesh family: in, dramatically fewer triangles, out, fast. When you're prototyping a workflow, testing whether a mesh is even worth working with at full density, or just need something light enough to preview, this is the node to grab.

The headline difference from the CGAL backend is how you tell it how much to cut. Fast Simplification thinks in fractions removed, not faces remaining:

• target_reduction - the fraction of faces to REMOVE. The tooltip is unambiguous: 0.5 = reduce to ~50% of original faces, 0.9 = reduce to ~10% of original. Default 0.5. If you're used to "target face count" dials, this is the mirror image of them.
• aggressiveness - how aggressively to simplify, 1–15, default 7. Higher means faster with lower quality; lower preserves detail at the cost of time. The default is a sane middle ground.

What comes out

decimated_mesh (the TRIMESH result) and info (a STRING with the actual outcome - worth a glance, because with a reduction-fraction dial you don't always get what you mentally aimed at). Feed a TRIMESH in from GeometryPack's loaders, run, and land the output on a viewer, saver, or the next node in the chain.

Installing it

Same pack install as everything here:

cd ComfyUI/custom_nodes
git clone https://github.com/PozzettiAndrea/ComfyUI-GeometryPack.git
cd ComfyUI-GeometryPack
pip install -r requirements.txt --upgrade
python install.py

Restart ComfyUI. ComfyUI Manager works too - search "GeometryPack" or use "Install via Git URL" with https://github.com/PozzettiAndrea/ComfyUI-GeometryPack.git. The pack's one-click installer is experimental (comfy-env, downloads pixi); the manual route is the reliable one per the README, and the pack is a fast-moving WIP, so keep it updated.

When to pick this one (and when not to)

Reach for Fast Simplification when you want a quick percentage cut with one dial and speed - it's great for knocking a dense scan down to something you can actually look at. Don't reach for it when you need a specific face count or maximum shape fidelity: that's the CGAL backend's job, and it will preserve volume and boundary far better for the same budget. And a real trap with this node specifically: because the reduction is a fraction, the same target_reduction on a 100k and a 10M mesh produces wildly different absolute counts. Read the info string before you trust it into an exporter.