CGFlowStudioGUIDES
Back to All Tutorials
Intermediate21 min read

Complete Hunyuan3D-2 ComfyUI Workflow Guide: Single-View, Multi-View & Turbo 3D Mesh Generation

Step-by-step tutorial on running Tencent's Hunyuan3D 2.0 natively in ComfyUI — covering Single-View, Multi-View, and Turbo 3D geometry mesh generation (.glb) and checkpoint setup.

By CGFlowStudio
•
2026-08-20

Complete Hunyuan3D-2 ComfyUI Workflow Guide: Single-View, Multi-View & Turbo 3D Mesh Generation

Tencent’s Hunyuan3D 2.0 represents a milestone in open-source 3D asset creation. Designed as a high-precision two-stage generative system, Hunyuan3D-2 converts 2D reference artwork and multi-view concept designs into detailed, clean 3D polygonal meshes (.glb) in seconds.

With official native support integrated directly into ComfyUI, generating 3D geometry is faster and more accessible than ever. In this tutorial guide, we walk you through setting up model checkpoints, selecting between Single-View, Multi-View (2mv), and Multi-View Turbo (2mv-turbo) workflows, isolating image backgrounds, and exporting high-precision 3D meshes to ComfyUI/output/mesh/.

1. Introduction & Current Native Capabilities

Before configuring your nodes, it is essential to understand how Hunyuan3D 2.0 is structured and what is currently supported inside native ComfyUI.

ℹ️
Important Note on Geometry vs. Texture Generation: Hunyuan3D 2.0 uses a two-stage pipeline: Stage 1 (Shape/Geometry Generation) and Stage 2 (Paint/Texture Generation). Native ComfyUI nodes currently focus on Stage 1 Geometry Mesh Generation (producing textured `.glb` shapes). Stage 2 texture refinement can be combined via community wrapper nodes.

Key Advantages of Hunyuan3D-2

  • High-Precision Geometry: Generates clean, manifold 3D meshes with crisp edges and organic surface curvature.
  • Multi-View Conditioning (2mv): Accepts front, left, back, and right reference views to eliminate 3D occlusion ambiguity.
  • Step-Distilled Turbo Variant (2mv-turbo): Enables rapid 3D geometry rendering at fixed CFG = 1.0.
  • Moderate VRAM Footprint: Runs comfortably on 12GB to 16GB consumer GPUs using FP16 weights.

2. Model Overview – Which Version Should You Use?

Hunyuan3D 2.0 provides three primary workflow variants inside ComfyUI:

Single-View 1 Image

Hunyuan3D-2 (Single)

Uses Hunyuan3Dv2Conditioning. Best for quick 3D generation from a single 2D concept image.

Multi-View Up to 4 Views

Hunyuan3D-2mv

Standard multi-view model using front, left, back, and right views for maximum 3D shape fidelity.

Turbo Fast Step-Distilled

Hunyuan3D-2mv-turbo

Step-distilled multi-view variant operating at CFG = 1.0 for fast iteration speed.

Model Comparison Table

Workflow Variant Input Requirements Generation Speed 3D Occlusion Accuracy
Hunyuan3D-2 (Single) 1 Reference Image Fast (~15–20s) Good for frontal views; predicts unseen back features.
Hunyuan3D-2mv Front (Mandatory) + Left/Back/Right Moderate (~30–45s) Best. Eliminates 3D blind spots via multi-angle views.
Hunyuan3D-2mv-turbo Front (Mandatory) + Optional Views Fastest (~8–12s) High precision at fixed CFG=1.0 distillation.

3. Getting Started – Loading Official Workflows

ComfyUI supports Hunyuan3D-2 via built-in template workflows and embedded PNG workflow files.

Loading via Template Library

  1. Update ComfyUI: Run git pull or update via ComfyUI Manager to register core nodes (Hunyuan3Dv2Conditioning, Hunyuan3Dv2MultiViewConditioning).
  2. Open Templates: Click Templates in the ComfyUI top menu and select 3D → Hunyuan3D-2.
  3. Drag & Drop PNG Workflows: Alternatively, drag any official Hunyuan3D-2 workflow PNG file directly onto your ComfyUI canvas to auto-load nodes.

4. Model Installation & Directory Structure

Download the Hunyuan3D 2.0 checkpoints from Tencent/Hunyuan3D-2 on Hugging Face and place them into your checkpoints directory:

📂 Hunyuan3D-2 Model Checkpoint Layout ComfyUI/models/checkpoints/
📂 ComfyUI/models/checkpoints/
📄 hunyuan3d-dit-v2-mv.safetensors
Standard Multi-View Checkpoint (Renamed from model.fp16.safetensors)
📄 hunyuan3d-dit-v2-mv-turbo.safetensors
Distilled Turbo Multi-View Checkpoint (CFG=1.0)
📄 hunyuan3d-dit-v2-single.safetensors
Single-View Checkpoint (For 1-image generation)
File Renaming Tip: Official Hugging Face downloads are named model.fp16.safetensors inside subfolders. Rename them to descriptive names (e.g. hunyuan3d-dit-v2-mv.safetensors) to avoid confusion inside the ComfyUI Checkpoint Loader menu.

5. Step-by-Step Workflow Walkthroughs

Workflow A: Hunyuan3D-2mv (Multi-View Master Setup)

The standard multi-view workflow uses up to four view images (front, left, back, right) to generate detailed 3D geometry.

Multi-View Routing Flow
Front View (Mandatory)
Left / Back / Right Views
Hunyuan3Dv2MultiViewConditioning
[View Angle Concatenation]
KSampler → SaveGLB Mesh Output

Step-by-Step Execution

  1. Load Checkpoint: Select hunyuan3d-dit-v2-mv.safetensors in the Checkpoint Loader node.
  2. Load Front View (Mandatory): Connect a clean, background-removed front-facing image into the front input socket of Hunyuan3Dv2MultiViewConditioning.
  3. Load Optional Views: Connect left, back, or right view images if available.
  4. Configure Sampler: Set steps to 30–50 and CFG to 3.0–5.0.
  5. Run Queue: Click Queue Prompt. Output mesh saves to ComfyUI/output/mesh/.

Workflow B: Hunyuan3D-2mv-turbo (Faster Multi-View)

The turbo variant utilizes step distillation to cut rendering times by 75%.

  1. Load Turbo Checkpoint: Select hunyuan3d-dit-v2-mv-turbo.safetensors.
  2. Lock CFG to 1.0: Ensure CFG scale is set strictly to 1.0 (increasing CFG on distilled models causes geometric deformation).
  3. Set Steps to 8–12: Sampling completes in under 10 seconds.

Workflow C: Hunyuan3D-2 Single-View Mode

Designed for fast 3D creation when only a single 2D concept image is available.

Single-View Pipeline Flow
Load Image (Front) → Hunyuan3Dv2Conditioning → KSampler → SaveGLB (.glb Mesh)
  1. Load Single Checkpoint: Select hunyuan3d-dit-v2-single.safetensors.
  2. Connect Conditioning Node: Route your image into Hunyuan3Dv2Conditioning instead of the MultiView node.
  3. Run Queue: Generates a complete 3D mesh predicting unseen rear details.

6. Practical Tips for Better 3D Results

✂️ Background Isolation is Essential

Always run your input images through a background removal node (such as BiRefNet or Rembg) before feeding them into Hunyuan3D conditioning nodes. Background noise will distort generated 3D mesh boundaries.

📂 Locating Exported .glb Files

Generated 3D meshes are automatically saved as .glb files under the ComfyUI/output/mesh/ directory. You can drag these directly into Blender, Windows 3D Viewer, or web viewers.

💡 Consistent Multi-View Lighting

When supplying front, left, back, and right images for multi-view mode, ensure neutral, flat lighting across all views to prevent lighting directional bias in the generated mesh.

7. Community Ecosystem & Texture Options

While core native ComfyUI support handles geometry mesh generation (Stage 1), creators seeking full PBR texture maps can explore community extension wrappers:

  • ComfyUI-Hunyuan3DWrapper: Community custom node package integrating Tencent’s Hunyuan3D-2-Paint model for secondary UV texture synthesis.
  • Tencent Official Repositories: Access source code, papers, and updates on the official Tencent Hunyuan3D-2 GitHub.

8. Conclusion & Next Steps

Hunyuan3D 2.0 provides local creators with a robust 3D geometry engine inside ComfyUI. By starting with Single-View generation for rapid prototyping and progressing to Multi-View (2mv) workflows for production assets, you can create clean 3D polygonal meshes directly from 2D artwork.

Next Steps to Explore

  • Drag your exported .glb files into Blender for UV unwrapping and sculpting.
  • Combine Hunyuan3D-2 mesh previews with TripoSplat for Gaussian point-cloud comparisons.
  • Browse more 3D guides on our ComfyHub Tutorials Directory.