Tripo: Smart low poly

A generated mesh can easily run tens of thousands of triangles. Great for a cinematic render, unusable for a game - engines want low counts, and VR wants them lower. TripoSmartLowPoly is the node that takes a high-poly mesh and intelligently reduces it down to a budget you set, while keeping the silhouette and (with bake on) the detail in the texture where it belongs.

This is the node the draft node's smart_low_poly flag secretly points at - turn that on during generation and you've run this pass before you've even downloaded the draft. As a standalone step, it's where you go when a mesh "works" but is too heavy for where it's going.

What it takes

Standard model_info in. The model_version is pinned to P-v2.0-20251225 (the production low-poly engine), so the interesting controls are elsewhere:

• face_limit - the budget, default 10000. This is the number you actually care about: set it to your game's target (say 5k for a hero asset, 1–2k for a background prop) and Tripo reduces toward it.
• quad - export quad topology instead of triangles. Quads are friendlier for subdivision workflows and sculpting in a DCC; triangles are what engines actually rasterize. Leave off unless you know you need quads.
• bake (on) - bakes the high-poly detail into normal/diffuse textures on the low-poly result. This is the thing that keeps a 2k mesh from looking like a 1999 game character. Turning it off gets you a faster, cleaner reduction with detail simply gone.
• part_names - if the mesh is segmented, apply the reduction to specific parts only.

Outputs are the standard pair: model_file (path to the reduced mesh in ComfyUI/output/) and updated model_info so it can continue into convert or texture.

Install

Same pack, same story:

cd ComfyUI/custom_nodes
git clone https://github.com/VAST-AI-Research/ComfyUI-Tripo
cd ComfyUI-Tripo && pip install -r requirements.txt

Or "Tripo for ComfyUI" in ComfyUI Manager. Key from platform.tripo3d.ai, and remember every run is a cloud task that spends credits.

The honest part

This is one of the most genuinely useful nodes in the pack, because it closes the "AI mesh into real asset" gap that most people bounce off - but understand what it is and isn't. It's smart decimation with texture baking, not a hand-made retopology. It'll get you an in-engine-ready LOD, and for many uses (props, background characters, anything not the camera's hero) that's all you need. For a character that's going to deform under animation, you'll still want to check the topology against your rig - aggressive reduction plus skinning can get crease-y. And note the face_limit ceiling here is 20000, so this isn't the node to "detail up" a mesh - if you need more faces than that, that's a refine job, not a low-poly job.