AI GitHub
TRELLIS for generating 3D assets
Generate a 3D representation from supported image or text conditioning workflows.
Key Takeaways
- TRELLIS uses a structured latent representation that can decode into multiple 3D output forms.
- Choose an output representation based on the next step in the project.
- Generate one object and inspect it from several angles.
- A plausible input view gives limited information about the unseen side.
How it works
TRELLIS uses a structured latent representation that can decode into multiple 3D output forms. The form you export affects what downstream tools can edit or render.
What you need
Choose an output representation based on the next step in the project. A renderable result and a mesh ready for editing or fabrication need different checks.
What to try first
Generate one object and inspect it from several angles. Export the representation your renderer expects and verify geometry as well as appearance.
Inspect the geometry under plain lighting and use an independent viewer. Textures or the original view can conceal holes and implausible shapes.
Where to look for failures
A plausible input view gives limited information about the unseen side. Generated geometry may need cleanup before animation, collision or fabrication.
Related Franklin pages
Decide whether the goal is to generate an object or reconstruct a captured scene. Inspect the asset from new viewpoints and verify what your renderer can actually consume.
- Train and inspect neural scene representations from captured views. Nerfstudio for reconstructing scenes from images
- Use differentiable Gaussian rasterization building blocks from Python. gsplat for Gaussian splatting research and rendering
For video assembled from authored HTML and motion graphics, see Franklin’s HyperFrames profile. HyperFrames
Project documentation
Try it for your workflow
Explore the project’s documentation for setup and usage, or save this page to a collection for later.
Read the project documentation ↗