Vi3W
Skip to content

Guides

ComfyUI 3D: how to generate 3D models in ComfyUI

ComfyUI 3D model generation explained: which node packs, which models (Hunyuan3D, TRELLIS, TripoSR, SF3D), a standard workflow, VRAM and errors.

2 min read

A network of glowing node boxes connected by curved cables, ending in a small 3D vase on a pedestal
Illustration. ComfyUI chains image generation, cleanup and 3D in one graph.

Quick answer

Can ComfyUI generate 3D models?

Yes. With community node packs, ComfyUI runs open image-to-3D models such as Hunyuan3D 2.x, TRELLIS, TRELLIS 2, TripoSR, Stable Fast 3D and InstantMesh, and saves the result as a GLB. The common workflow is image in, background removed, 3D shape generated, texture baked, GLB out, with optional text-to-image in front.

How 3D works in ComfyUI

ComfyUI started as an image tool, but its node graph suits 3D pipelines well. Community node packs add:

  • 3D generation nodes for specific models.
  • Mesh nodes: decimate, fill holes, UV unwrap, bake textures.
  • 3D preview inside ComfyUI.
  • Save nodes for GLB, OBJ and similar.

Node packs to know

  • ComfyUI-3D-Pack (MrForExample, MIT): a big suite covering Hunyuan3D 2.1 and 2.0, TRELLIS, Stable Fast 3D, TripoSR, InstantMesh, Unique3D and more, with example workflows. Its README targets Windows 10/11, Python 3.12 and CUDA 12.4. ComfyUI 3D Pack
  • ComfyUI-Hunyuan3DWrapper (kijai): focused Hunyuan3D nodes. Hunyuan 3D in ComfyUI
  • TRELLIS 2 node packs: community packs, including low-VRAM variants. TRELLIS 2 in ComfyUI
A chain of glowing nodes ending in a textured mushroom
Illustration. Each node is one step; the graph is your pipeline.

A standard ComfyUI 3D workflow

  1. (Optional) Text to image: generate the object on a plain background.
  2. Load Image and remove background.
  3. 3D generation with your chosen model.
  4. Mesh cleanup: decimate, remove floaters, fill holes.
  5. Texture generation or bake.
  6. Save GLB.
  7. Preview in the 3D viewer node.

Choosing a model by GPU

  • 6 GB: TripoSR, Stable Fast 3D, Hunyuan3D shape-only.
  • 16 GB: TRELLIS, Hunyuan3D 2.0 with texture.
  • 24 GB+: TRELLIS 2; Hunyuan3D 2.1 with texture (about 29 GB, or low-VRAM mode).

From each model's README. More: GPU for AI 3D generation.

Common problems

  • Compiled dependencies: 3D packs often need custom CUDA extensions built for your exact Python, PyTorch and CUDA versions. Follow the pack's install script.
  • Version drift: update ComfyUI and node packs together.
  • 3D preview not working: usually a browser or WebGL issue; the GLB file is still saved.

Export to Blender and engines

Import the GLB into Blender with File ‣ Import ‣ glTF 2.0, then retopologise for games. See AI 3D model generator for Blender.

The managed alternative

If you'd rather not maintain a ComfyUI install, Vi3W runs Hunyuan 3D, TRELLIS and TRELLIS 2 plus commercial models (Meshy, Tripo, Rodin) in the browser, with retopology and GLB, FBX or OBJ export. Try Vi3W.

Prompt-to-3D pipelines are covered in open source text to 3D.

Frequently asked questions

comfyui 3d model workflow

Start from a node pack's example workflow: Load Image, remove background, model's 3D sampler, mesh decode, texture, Save GLB. Add a text-to-image stage in front for prompt-to-3D.

comfyui image to 3d free

ComfyUI and the node packs are free; the models are free to download. Check each model's licence for commercial use.

comfyui best 3d model generator

Most people use Hunyuan3D 2.x or TRELLIS 2 for quality, and TripoSR or Stable Fast 3D for speed on smaller GPUs.

Last reviewed