Choose theme
AI Visuals
TRELLIS.2
TRELLIS.2 is a Microsoft model that generates a textured 3D asset from a single image, including physically based rendering materials.
The 4B-parameter model comes with pretrained weights, inference examples and training code. Its O-Voxel representation stores geometry and material attributes together. Use this as a first read, not a recommendation. Open the original project before trusting details like terms, limits, privacy, cost, setup, or safety.
What it is
Image-to-3D generation model
TRELLIS.2 is positioned as a large 3D generative model for turning images into textured 3D assets, with code paths for inference, texture generation, training, and exported GLB assets.
Why it stands out
O-Voxel and PBR material focus
O-Voxel stores geometry and appearance together. The exported assets include physically based rendering materials.
Availability
Public repo with weights and demos
The repository includes setup instructions, example scripts, web demo files, Hugging Face pretrained-weight links, data-preparation guidance, and training code for readers who want to inspect the workflow.
Why it matters
What makes it useful
The image-to-3D example turns a reference image into a textured mesh and exports a GLB for a 3D workflow. It also renders a video of the asset under environmental lighting, giving a builder several views to examine before using the generated object.
What to know
Where it fits
There are two starting points: generate an asset from an image, or supply an existing mesh and a reference image to the separate texturing example. The second path adds PBR textures to a given shape, so a builder with geometry already prepared can try the material stage without starting from image-to-3D generation.
Notable points
What stands out
Open surfaces such as clothing and leaves are among the cases Microsoft names for O-Voxel. The representation also covers non-manifold geometry and enclosed internal structures. Material attributes include base color, roughness, metallic values and opacity.
Before using
What to review
The repository says the code is tested only on Linux and requires an NVIDIA GPU with at least 24GB of memory; it names A100 and H100 as verified hardware.
Follow the published CUDA and dependency instructions for the chosen environment. The examples require a local model setup rather than just a browser upload.
The model card notes that raw meshes may contain small holes or topological discontinuities. Inspect the resulting geometry for the job you intend to use it for.
Check permission to use the input image and the generated asset for its intended purpose, alongside the current model terms. The repository notes that some dependencies have separate license terms.
The published pipeline names Meta's DINOv3 ViT-L/16 checkpoint as a dependency. Its Hugging Face page requires agreeing to share contact information for access; check that prerequisite before planning the local setup.
One export detail matters if your asset needs transparency: the repository says the GLB starts in OPAQUE mode even though its texture keeps an alpha channel. The repository instructs readers to connect that channel to the material's opacity or alpha input after importing the asset into their 3D software. A preserved alpha texture alone does not make the imported material transparent.
Reader fit
Who may find it relevant
3D builders trying generated props, textured assets or material experiments.
Researchers who want the training code and O-Voxel data-preparation workflow.
Readers comfortable with Python and a GPU environment; the repository examples are more involved than a lightweight desktop utility.
Editorial note
Why LifeHubber lists it
A builder exploring thin, open structures such as leaves or clothing has a reason to consider how a 3D model represents a shape. Microsoft describes O-Voxel as avoiding the closed-surface constraint of iso-surface fields while storing geometry and materials together. That makes TRELLIS.2 worth exploring for assets whose open surfaces or internal structures matter to the design; the generated mesh still needs inspection.
Source links
Source materials
Reader note
Before relying on this entry
LifeHubber lists entries to help readers inspect AI projects, not to endorse them or prove they are safe, suitable, accurate, maintained, or right for a specific use. We do not verify every entry in depth. Before relying on anything listed, review the original materials, terms, privacy practices, limits, and risks that matter for your situation.
More in AI Models
Keep browsing this category
Explore more AI model resources.
Gemma 4
google/gemma-4
A Google DeepMind Gemma 4 model family collection with public checkpoints including Gemma 4 12B, a dense multimodal model Google describes around local agentic workflows, native audio input, and encoder-free vision/audio handling.
MiniMax H3 Integrations
MiniMax-AI/awesome-minimax-h3-integration
A community-maintained MiniMax H3 integration index that maps checkpoints, hardware and VRAM starting points, runtimes, ComfyUI nodes, prompting tools, acceleration routes, and deployment options.
Isaac GR00T N1.7
nvidia/gr00t-n17
An NVIDIA Isaac GR00T N1.7 vision-language-action model family for humanoid and generalist robot skills, with a 3B model, post-trained variants, GitHub code, inference and fine-tuning notes, LeRobot-format workflow support, and official robotics developer materials.
For project maintainers
Listed here? You can use the badge.
If you maintain a project with a current LifeHubber listing, you may add the optional “Listed on LifeHubber AI Resources” badge to its README, docs, or website. No introduction or permission request is needed.
