Hunyuan3D: Smart Topology

AI-generated 3D meshes are catastrophically overbuilt. Hunyuan3D will cheerfully hand you a model with half a million to a million and a half faces - fine for a render, useless for a game, and a pain to rig, deform, or edit. Hunyuan3D: Smart Topology is the cleanup crew: a cloud retopology pass that rebuilds your high-poly blob into a clean, low-poly mesh with sensible edge flow instead of a voxel-esque tri soup.

Like every node in the partner/3d family, it's built into ComfyUI core - no extension, no install. The actual work happens on Tencent's servers: you log in with a ComfyUI Account, prepaid credits, and the node uploads your mesh, runs a job, and polls until it's done. The price badge on the node tells you what you're about to spend before you commit. Fair warning: this is one of the most expensive nodes in the whole partner/3d group, so it's worth using deliberately rather than reflexively.

The inputs that matter

• model_3d - a GLB or OBJ (the only formats supported). Max 200 MB, and the docs recommend it specifically for high-poly input, which is exactly the case where it earns its keep.
• polygon_type - triangle or quadrilateral. Quads are what you want if you're heading into Blender or ZBrush for further editing or rigging; triangles are more engine-neutral.
• face_level - low, medium, or high. This is your polygon-reduction level. Low gets you a game-ready budget, high keeps more detail for sculpting territory.
• seed - per the tooltip, just controls whether the node re-runs. Results are non-deterministic regardless of seed, so don't treat it like a reproducibility knob.

The single output is an OBJ - the cleaned-up mesh. That's it. Feed it into Save 3D Model, Preview 3D & Animation, or straight to whatever you're building in.

Where it fits

The natural pipeline is: generate with Hunyuan3D: Text to Model at a high face_count, then Smart Topology it down to something usable. It also pairs well before Hunyuan3D: Model to UV, since UV unfolding has a strict 30,000-face ceiling - retopologizing first gets you under that limit and gives the unwrapper clean geometry to work with.

What to watch out for

Two things bite people. First, format support is GLB/OBJ only - if your source is an FBX or STL, convert it before this node. Second, the cost: at this node's price per run, you don't want to be experimenting. Decide on your face_level up front and run it once.

The honest take: Blender's decimate modifier can get you 80% of the way there for free. Smart Topology earns its credits when you need good topology - quads where it counts, clean flow for deformation - not just fewer triangles. For game assets and rigged characters, that difference is the whole ballgame.