Vi3W
Skip to content

Guides

Hunyuan 3D ComfyUI workflow: from image to textured 3D model in nodes

Hunyuan 3D ComfyUI workflow: install the wrapper nodes, load weights, build image-to-3D with texture, fix VRAM and compile errors, export GLB.

2 min read

A ceramic frog shown three ways: textured, as a plain grey mesh with wireframe, and as its unwrapped texture sheet
Illustration. A ComfyUI Hunyuan workflow builds shape, then texture.

Quick answer

How do I use Hunyuan 3D in ComfyUI?

Install a Hunyuan3D node pack, such as kijai's ComfyUI-Hunyuan3DWrapper, through ComfyUI Manager, download the model weights, and load an example workflow: image in, background removal, shape generation, mesh clean-up, then texture generation and GLB export. Shape alone needs far less VRAM than shape plus texture.

Why ComfyUI for Hunyuan 3D

ComfyUI lets you chain steps visually: generate an image, cut out the background, build a 3D shape, texture it and save a GLB, all in one graph you can reuse. For many people it's easier than the raw repo, and it's the natural home for text-to-image-to-3D pipelines.

Video: ​​Run Hunyuan 3D 2.1 in ComfyUI – Fast, High-Fidelity Image-to-3D Workflow (Blender + Multiview), by ComfyUI Workflow Blog (YouTube)

Install

  1. Install ComfyUI and ComfyUI Manager.
  2. Install a Hunyuan3D node pack. A widely used one is kijai/ComfyUI-Hunyuan3DWrapper. Check its README for supported Hunyuan versions and any compiled parts needed for texturing.
  3. Download weights from Tencent's Hugging Face pages (for example tencent/Hunyuan3D-2 or tencent/Hunyuan3D-2.1) into the folder the node pack expects.
  4. Restart ComfyUI and load an example workflow from the node pack.

Using the weights means accepting Tencent's community licence, which doesn't apply in the EU, UK or South Korea.

A typical Hunyuan 3D ComfyUI workflow

  1. Load Image: one object, plain background.
  2. Remove background with a segmentation node.
  3. Shape generation with the Hunyuan3D sampler (choose steps and guidance).
  4. Decode to mesh, then optional decimation or clean-up nodes.
  5. Texture generation with the paint model (multi-view render, paint, bake).
  6. Save as GLB.

For multiple views, use the 2mv model and feed front, side and back images.

A frog shown textured, as wireframe and as its texture sheet
Illustration. Shape, texture and UV layout are separate stages in the graph.

VRAM and settings

From Tencent's READMEs: Hunyuan3D 2.0 needs about 6 GB for shape and 16 GB with texture; 2.1 needs 10 GB for shape and 29 GB for both. If you run out:

  • Run shape only first; texture in a second pass.
  • Use 2mini, or the Turbo/Fast variants.
  • Lower octree resolution or step counts.

Common errors

  • Missing custom rasterizer / renderer: texturing needs compiled extensions; build them for your exact Python and CUDA versions.
  • "detail not found" or model-load errors: weights are in the wrong folder or are the wrong version for the node.
  • Black or blank textures: usually a failed render step; check the console.

Into Blender

Import the GLB with File ‣ Import ‣ glTF 2.0, then clean up: Hunyuan 3D to Blender.

Other 3D nodes for ComfyUI

TRELLIS 2, TripoSR and more: ComfyUI 3D and ComfyUI 3D Pack.

The no-ComfyUI route

Vi3W runs Hunyuan 3D 2.1, 2, 2 Mini and hosted 3.x models in the browser with no install, plus TRELLIS, Meshy and Tripo, and exports a retopologised GLB, FBX or OBJ. Try Vi3W.

Frequently asked questions

hunyuan 3d comfyui workflow

A typical workflow: Load Image, remove background, Hunyuan3D shape sampler, decode to mesh, decimate or clean, then the texture pipeline and a Save GLB node. Start from the node pack's example workflows.

comfyui hunyuan 3d with texture

Texturing needs the texture model and its compiled renderer extensions, plus much more VRAM than shape. If it fails, generate shape only and texture separately.

hunyuan 3d comfyui kijai

kijai/ComfyUI-Hunyuan3DWrapper is a widely used community node pack for Hunyuan3D in ComfyUI. Follow its README for the compiled components needed for texturing.

Last reviewed