← Constellation
3D & Gaming/3D Generation

TRELLIS

Open-source image-to-3D model. Generate full 3D meshes from a single image, free and self-hostable.

Visit trellis3d.github.io

External link. Not endorsed — curated for usefulness.

What is TRELLIS?

TRELLIS is an open-source image-to-3D generation tool that creates full 3D meshes from single 2D images, developed as a research project and available for free use and self-hosting.

The tool uses machine learning to reconstruct three-dimensional models from photographs or digital images without requiring multiple input views or manual intervention. Users upload an image, and the system generates a complete 3D mesh suitable for rendering, editing, or integration into 3D applications. TRELLIS operates entirely in the browser or can be deployed locally, eliminating dependency on cloud servers and ensuring data privacy. The open-source nature allows developers and researchers to inspect the codebase, modify the model architecture, and adapt the tool for specialized use cases. Generated meshes are typically in standard formats like OBJ or GLB, making them compatible with Blender, Three.js, Babylon.js, and other 3D software ecosystems.

The tool serves several user groups: game developers prototyping assets quickly, 3D artists automating modeling workflows, researchers in computer vision, and hobbyists creating digital content. It requires no subscription or authentication, removing barriers to experimentation and large-scale batch processing. Performance depends on local hardware when self-hosted, allowing users with GPUs to accelerate generation. While the free model provides reasonable quality for many applications, results vary based on image complexity, lighting conditions, and subject matter—simple objects and clear photography typically produce cleaner reconstructions than cluttered scenes or extreme angles.

TRELLIS lacks some advanced features found in commercial competitors: it doesn't generate textures or detailed materials by default, doesn't handle transparent or reflective surfaces as robustly, and produces lower polygon counts than some specialized tools. The codebase is maintained on GitHub, with periodic updates reflecting improvements in under

Submitted by Editorial roundup