Summary

This ComfyUI wrapper integrates Microsoft's TRELLIS.2 for native, compact structured latents in 3D generation. It targets ComfyUI users seeking advanced 3D asset creation, reconstruction, and texturing, streamlining complex 3D pipelines.

How It Works

The project embeds TRELLIS.2 functionalities into ComfyUI nodes, enabling structured latent representations for 3D generation. It offers nodes for multi-view processing, voxel/sparse/shape mesh generation, projection mapping, hole filling, texturing, and reconstruction, employing advanced 3D asset manipulation techniques.

Quick Start & Requirements

Installation involves cloning into ComfyUI's custom nodes. Prerequisites include facebook/dinov3 models and potentially natten. Pre-compiled wheels are provided for Windows with specific Python (3.11, 3.13) and Torch (2.7.0, 2.8.0, 2.10.0) versions, requiring CUDA (12.8, 13.1). Custom builds for o_voxel, Cumesh, FlexGEMM are available. Installation via requirements.txt or wheels is supported. Links to TRELLIS.2 and dependencies are provided.

Highlighted Details

• Extensive node additions since January 2026: multi-view projection, sparse/voxel mesh generation, advanced hole filling, mesh reconstruction, multi-view texturing.
• Experimental features: Sparse Structure Resolution up to 128, RK4/RK5 samplers, experimental multi-view texturing.
• Support for Pixal3D-T, FP8 models, sdpa, flash_attn_3 backends.
• Nodes for image preprocessing (incl. background removal) and video data extraction.
• "DINO-Lock" functionality, mesh simplification/smoothing.

Maintenance & Community

Active development with frequent updates (April-May 2026). A Discord community is noted, alongside a "Blackwell Fix" repository, indicating community engagement and issue resolution.

Licensing & Compatibility

The README omits license information, creating ambiguity for usage rights and commercial compatibility.

Limitations & Caveats

Primarily Windows-focused, with unclear Linux/macOS support. Pixal3D-T compatibility is not universal. Experimental features may be unstable. Custom dependency builds suggest potential integration challenges or specific hardware needs.