Microsoft's TRELLIS 2, a 4-billion-parameter 3D generation model released in December 2025, has become the dominant open-source tool for converting 2D images into textured 3D models. Available free under the MIT license on Hugging Face, the model runs locally inside ComfyUI, a node-based visual interface, and produces production-ready 3D assets complete with physically based rendering (PBR) textures in seconds. Unlike competing commercial tools like Meshy and Tripo that charge per generation, TRELLIS 2 allows unlimited free generations on any capable NVIDIA GPU.

What Makes TRELLIS 2 Stand Out From Other 3D AI Tools?

TRELLIS 2 has overtaken every other open-source 3D option in the ComfyUI ecosystem because it delivers texture fidelity and commercial-use freedom that competitors cannot match. The model accepts a single image or text prompt and outputs high-fidelity 3D assets with transparency support and hard-surface detail, exporting to multiple formats including GLB, OBJ, and STL. Community developers have optimized quantized versions that run on GPUs with as little as 6 gigabytes of video memory, down from the original 24-gigabyte requirement, making the tool accessible to creators with mid-range hardware like an RTX 2060 or GTX 1080.

The competitive landscape reveals why TRELLIS 2 has gained such rapid adoption. Tencent's Hunyuan3D is also open-source but historically trails on texture quality. Meshy and Tripo offer polished interfaces and specialized features like low-poly optimization and dual rendering modes, but both operate as commercial APIs that bill per generation. TRELLIS 2 is the only top-tier option that combines open-source availability, commercial-use licensing, and production-grade output quality without per-generation fees.

How to Set Up TRELLIS 2 in ComfyUI for Your First 3D Generation?

• Install the Foundation: Download ComfyUI and the ComfyUI Manager, which provides a searchable interface for custom nodes and workflows without manual configuration.
• Add the TRELLIS 2 Node: Search for and install the ComfyUI-TRELLIS2 custom node through the Manager, which now ships native TRELLIS 2 nodes as part of the standard ecosystem.
• Build Your Workflow: Connect nodes in sequence: load or generate a source image, feed it into the TRELLIS Mesh Generator node, pass the geometry to the TRELLIS Texture Generator node to bake PBR materials, then export the final GLB file ready for Blender, Unity, Unreal Engine, or 3D printing.
• Prepare Your Input: Use a clean PNG image with the background removed for best results, as this improves mesh geometry and texture accuracy.
• Queue and Export: Submit your workflow to the processing queue and download the production-ready 3D asset in your preferred format.

Running TRELLIS 2 inside ComfyUI offers significant advantages over the standalone command-line interface. The node-based visual workflow enables drag-and-drop pipeline chaining, automatic batch processing through the queue system, and live preview integration with Stable Diffusion image generation on the same canvas. This unified environment allows creators to generate a 2D image, immediately convert it to 3D geometry, apply textures, and export a finished asset without switching between separate tools or writing scripts.

What Hardware Do You Actually Need to Run TRELLIS 2 Locally?

The hardware requirements have dropped dramatically thanks to community optimization efforts. The minimum specification is an NVIDIA GPU with 6 gigabytes of video memory running a quantized version of the model, which requires approximately 10 gigabytes of storage for model weights and dependencies on an SSD. The recommended setup uses an NVIDIA GPU with 16 gigabytes or more of VRAM, such as an RTX 4070 or higher, for faster inference and larger batch processing. For creators whose hardware cannot meet even the quantized requirements, cloud-based implementations of TRELLIS 2 run on optimized infrastructure without requiring local installation.

The model's architecture uses a Structured Latent (SLAT) transformer with flow-matching technology to achieve 16-times spatial compression, enabling output resolutions up to 1536 cubic voxels of PBR-textured 3D geometry. This technical efficiency is why community developers successfully reduced the VRAM footprint from 24 gigabytes to 8 gigabytes, making the tool practical for creators using consumer-grade graphics cards from the GTX 1080 and RTX 2070 generation.

TRELLIS 2 represents a significant shift in how open-source 3D generation is distributed and accessed. By hosting the model weights on Hugging Face under a permissive MIT license, Microsoft has enabled a global community of developers to build optimizations, integrations, and workflows that lower barriers to entry for game developers, 3D artists, and product designers. The rapid adoption inside ComfyUI demonstrates that creators prioritize both output quality and cost-free, unlimited usage over polished commercial interfaces when the technical quality is competitive.