TRELLIS

What is TRELLIS?

TRELLIS is Microsoft Research's open-source model for generating 3D assets from a single image or a text description. The name comes from the paper 'TRELLIS: Structured 3D Latents for Scalable and Versatile 3D Generation', and the key idea is a representation called SLAT that stores shape information as a combination of points and sparse grids.

That representation lets the model produce cleaner geometry with more detail than earlier image-to-3D approaches, and it does it quickly. On a modern RTX GPU you can generate a textured mesh in seconds, then export it as glTF or another common format.

Key Features

• Image-to-3D and text-to-3D generation
• SLAT structured latent representation for detailed geometry
• Automatic texture generation on the mesh
• Fast inference on consumer GPUs
• Fully open source under the MIT license
• Gradio demo for easy testing
• Exports to standard mesh formats like glTF

How to get started

Start by cloning the GitHub repository and installing the requirements into a Python environment. The project also has a Gradio demo, so once the dependencies are in place you can launch a friendly web interface and drag in an image.

Model weights download automatically on first use or via a provided script, and generation runs locally on your GPU. A machine with around 8 GB or more of VRAM handles the standard model, while the higher quality variants want a bigger card.

• Clone the repository from GitHub
• Set up a Python environment and install dependencies
• Download the model weights
• Launch the Gradio demo or run the CLI
• Feed in an image or prompt and export the result

Use cases

• Generating game assets and props from concept art
• Creating 3D content for research and education
• Prototyping product visualizations from photos
• Building 3D datasets for model training
• Learning how modern 3D diffusion models work

Pricing and licensing

TRELLIS is completely free and open source under the MIT license, so you can use it commercially, modify the code and even integrate it into your own tools without paying licensing fees.

Your only real cost is hardware: you need a GPU with enough VRAM, and electricity for the inference runs. That makes it one of the most budget-friendly ways to do high-quality AI 3D generation.

Pros and cons

The biggest plus is that you own your pipeline. No subscriptions, no credit meters, no data leaving your machine, and the quality is competitive with commercial services, particularly for clean single-object generation.

The trade-offs are setup effort and control. You need a capable GPU and some patience installing dependencies, and like all AI generators the results occasionally need cleanup before they are production ready.

Alternatives

• Hunyuan3D: Tencent's open-source 3D generation models
• Tripo: commercial text-to-3D and image-to-3D API
• Meshy: browser-based AI 3D generation
• CSM: image-to-3D pipeline with an API
• Stable Fast 3D: fast single-image 3D from Stability AI