visualbruno/ComfyUI-Trellis2

What it solves

This project provides a ComfyUI wrapper for TRELLIS.2, allowing users to integrate 3D asset generation from images into a node-based visual workflow. It simplifies the process of creating 3D meshes, texturing them, and refining them using various specialized nodes.

How it works

It implements the series of steps required by the TRELLIS.2 framework to transform images into 3D objects. The wrapper provides a suite of nodes for:

• Generation: Creating multi-view images or voxels from a single image (e.g., using Sparse, ImageCond, or Shape Cascade generators).
• Mesh Creation: Converting voxels to meshes and reconstructing meshes using various algorithms (e.g., Voxel to Mesh, Trellis2ReconstructMesh).
• Mesh Refinement: Cleaning up the geometry with tools for smoothing, welding vertices, simplifying meshes, and filling holes using libraries like PyMeshlab and Meshlib.
• Texturing: Applying textures to the 3D model based on multi-view images.

Who it’s for

Artists, 3D designers, and AI researchers who use ComfyUI and want to generate 3D assets from 2D images using the TRELLIS.2 architecture.

Highlights

• Comprehensive Mesh Toolset: Includes nodes for smoothing, simplifying, and watertight mesh creation (Voxel to Mesh).
• LMD Support: Support for various models including Pixal3D-T and DINOv3 features extractor.
• Flexible Generation: Multiple generator types (Sparse, ImageCond, Shape) for different 3D reconstruction needs.
• Multiview Rendering: Integrated rendering capabilities via Nvdiffrast.
• Advanced Post-Processing: Specialized nodes for hole filling, vertex welding, and mesh simplification.