Skip to content

TRELLIS

Generate 3D assets from text or image prompts

Visit official site

What TRELLIS is

The model is described as a native 3D generation model built on a unified structured latent representation and a rectified flow transformer. It combines sparse 3D grids with dense multi-view visual features to capture both geometry and appearance.

What you can do with it

  • Generate 3D assets from text or image conditions
  • Choose the output format for downstream use
  • Apply local 3D editing to generated assets
  • Create assets with detailed geometry and textures

Model notes

  • Structured latent representation decodes into several output formats
  • Formats include radiance fields, 3D Gaussians and meshes
  • Dense multi-view features are drawn from a vision foundation model
  • A rectified flow transformer is customized for the structured latents

Getting started

  • Visit the project site or the GitHub page
  • Download and install the code and models
  • Set up a 3D generation environment from the guides
  • Generate 3D assets using text or image prompts

Pros & cons

✓ What we like

  • Output format can be chosen for different downstream needs
  • Local 3D editing is listed among the capabilities

! What to watch out for

  • Code, models and data are stated as forthcoming releases
  • Generation is conditioned on text or image inputs

FAQ

What outputs can it produce?

The material lists radiance fields, 3D Gaussians and meshes as decodable formats.

What conditions generation?

Text or image conditions are described as the inputs for generating 3D assets.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review