Hy3D_2_1SimpleMeshGen

Most of this pack splits shape generation into stages: sample the latent, decode the latent, and so on. Hy3D_2_1SimpleMeshGen is the shortcut - a single node that takes an image and returns a finished trimesh, end to end. It's the "I just want a mesh, now" button, and it's built around the 2.1 shape model rather than the original 2.0 pipeline.

Under the hood it's doing everything the big workflow does: it loads the shape DiT with the 2.1 config (dit_config_2_1.yaml), removes the background from your input with an internal rembg pass, runs the flow-matching sampler at your chosen octree_resolution, and extracts the mesh - all in one call, returning a trimesh you can immediately render, remesh, or export.

Inputs

• model - the shape model checkpoint, loaded from ComfyUI/models/diffusion_models/. This is the 2.1 weights file (the tooltip points you at the diffusion_models folder; you want the Hunyuan3D-2.1 shape safetensors, not the 2.0 one - the node's config expects the 2.1 architecture).
• image - your input image. It runs background removal internally if the image is RGB, so you can feed it an unprocessed render and it'll clean it up.
• steps (default 50) - diffusion steps.
• guidance_scale (default 5.0) - CFG strength.
• octree_resolution (default 384, 32–1024 in steps of 32) - the resolution of the voxel grid the mesh is extracted from. This is the quality/memory knob: 384 is the sensible default; push toward 512+ for detail-hungry hard-surface objects if your VRAM allows, and drop it for quick drafts.

Output: a single trimesh (TRIMESH).

The honest tradeoff

"Simple" here means "fewer moving parts," not "worse." What you give up is control and the intermediate artifacts: with the staged pipeline you can peek at the raw latent, tune the VAE decode, and re-run just the decode without re-sampling. This node bakes all of that into one shot - if the geometry is wrong, you re-sample from scratch. For most people that's the right trade, because the default decode settings in the staged path are exactly what this node uses anyway.

Where it fits: straight into the texturing side of the pack. Take the trimesh from here → Hy3DMeshUVWrap → Hy3DRenderMultiView → Hy3DSampleMultiView (paint) → bake → apply → Hy3DExportMesh. Or skip the paint entirely and just Preview3D/export the raw shape if you only needed geometry.

Notes

Hunyuan3D 2.1 carries Tencent's territory-limited community license - no grant for the EU, UK, or South Korea - so read model-licensing.md before you ship something commercial on it. Install is the pack install (ComfyUI Manager, search "Hunyuan3DWrapper", or clone into custom_nodes and pip install -r requirements.txt), and it needs the 2.1 shape model downloaded into diffusion_models/ on your own - this node won't fetch it for you.