This repository provides an extensive suite of custom nodes for ComfyUI, enabling users to generate and process 3D assets using cutting-edge algorithms and models. It targets users interested in 3D content creation, AI-driven asset generation, and integrating advanced 3D techniques into their workflows, offering a unified platform for complex 3D pipelines.

How It Works

The pack integrates various state-of-the-art 3D generation and reconstruction models, including NeRF, 3D Gaussian Splatting (3DGS), and multi-view diffusion techniques. It provides nodes for converting single images or multi-view inputs into 3D meshes with textures, generating novel views, and visualizing 3D assets directly within ComfyUI using libraries like gsplat.js and three.js. This approach allows for a modular and composable workflow, leveraging ComfyUI's node-based interface for complex 3D pipelines.

Quick Start & Requirements

• Installation: Installable via ComfyUI-Manager or manual clone. Pre-built binaries are available for Windows 10/11 with Python 3.12, CUDA 12.4, and PyTorch 2.5.1+cu124.
• Prerequisites: Visual Studio Build Tools (Windows) or GCC/G++ (Linux) are required for specific JIT-compiled nodes. Manual model downloads may be necessary if automatic downloads fail.
• Resources: Requires significant VRAM for certain models (e.g., 16GB+ for Era3D).
• Documentation: Example Workflows are available.

Highlighted Details

• Integrates a wide array of models: TRELLIS, Hunyuan3D (v1 & v2), StableFast3D, CharacterGen, Unique3D, Era3D, InstantMesh, Zero123++, CRM, TripoSR, Wonder3D, LGM, Triplane Gaussian Transformers, FlexiCubes, Instant NGP.
• Features direct 3DGS and 3D Mesh visualization within ComfyUI.
• Includes nodes for generating orbit camera poses and baking multi-view images into UV textures.
• Supports multiple 3D file formats for import/export (.obj, .ply, .glb).

Maintenance & Community

• Actively developed with a clear roadmap for future improvements, including better 3DGS/NeRF to Mesh conversion and integration of new MVS algorithms.
• Contribution guidelines are provided.

Licensing & Compatibility

• The repository itself appears to be under a permissive license, but it integrates models with various licenses (e.g., StabilityAI's terms of usage for StableFast3D). Users must ensure compatibility with their intended use cases, especially for commercial applications.

Limitations & Caveats

• Some nodes rely on JIT-compiled C++ extensions requiring specific build tools, which can complicate setup.
• Certain models have high VRAM requirements, potentially limiting accessibility on lower-end hardware.
• Users must manage model downloads and placement, as automatic downloads may not always succeed.