Practical overview
When it fits
Use Trellis2.C when a 3D asset workflow calls for local image-to-3D generation or mesh processing, and you can provide the required model packages and build dependencies.
Who it is for
- 3D artists and asset creators generating models from reference images
- Developers preparing existing meshes with generated materials or part segmentation
- Users who want local inference and can manage model packages and a native build environment
Documented uses
- Generate a textured GLB/glTF asset from a single image.
- Create PBR materials for an existing triangle mesh using a reference image.
- Separate a static GLB/glTF model into parts.
- Run automatic rigging on assets.
Requirements & limits
The repository specifies an MIT license and documents source builds for Linux and Windows. The CUDA build requires the CUDA Toolkit; both documented inference backend builds require the Vulkan SDK. Model packages must be available locally.
- The toolkit is a locally built application rather than a hosted generation workflow.
- Both CUDA and Vulkan backends are documented; backend selection is made at build time.
- Image-to-3D pipelines and mesh processing require locally prepared model packages.
- Mesh texturing creates a new static GLB and does not preserve several source-asset properties, including hierarchy, UVs, skins and animations.
- The README documents separate image-generation pipelines with different model families and resolution profiles.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Build the Linux CUDA backend
Configure and compile Trellis2.C for Linux with its CUDA inference backend.
Before you start
- Git and CMake 3.22 or newer
- GCC or Clang
- Vulkan SDK, including loader, headers and glslc
- CUDA Toolkit
Steps
Clone the repository with its submodules using the command below.
git clone --recursive https://github.com/Wimacs/trellis2.c.gitFrom the repository directory, configure with
cmake -S . -B build-cuda -DCMAKE_BUILD_TYPE=Release -DTRELLIS2_C_BACKEND=cuda.Compile with
cmake --build build-cuda -j.
Expected outcome
A Linux build configured to use the CUDA backend.
Guide limitations
- This is a build procedure; model packages and pipeline execution are separate steps.
- The Vulkan SDK is required even for this backend because pipeline tools also build Vulkan mesh-processing components.
- CUDA defaults to compute capability 8.9; another GPU generation can be selected with CMAKE_CUDA_ARCHITECTURES.
Generate a textured GLB from one image
Run the TRELLIS.2 image-to-3D pipeline to create a textured 3D asset.
Before you start
- A built executable, such as
./build-cuda/trellis2-image-to-gltf - Locally available TRELLIS.2 and DINOv3 model packages
- An input image
Steps
Ensure the required model packages have been downloaded or converted and are locally accessible.
Run
./build-cuda/trellis2-image-to-gltf --model ../TRELLIS.2/TRELLIS.2-4B --dino ../TRELLIS.2/dinov3-vitl16-pretrain-lvd1689m --image example_image/T.png --pipeline 1024 --output output.glb.
Expected outcome
The pipeline writes a textured GLB/glTF asset to the specified output path.
Guide limitations
- Opaque input uses an auto-discovered BiRefNet checkpoint when available; transparent RGBA input is used directly.
- Model packages must already be available locally.
- The README describes the command and expected output but does not establish execution results for a particular machine.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 16 October 2026
- Original website title
- Wimacs/trellis2.c · Official homepage title · retained in its source language