TRELLIS 2
43
Overview
- Create production-ready 3D assets from images instantly with fully rendering-free and optimization-free conversion, using simple pre- and post-processing methods.
- Generate models with complex structures like open surfaces and non-manifold geometry that other methods can't handle, using the flexible O-Voxel structure.
- Produce PBR-ready assets with accurate surface attributes like Base Color, Roughness, and Metallic for photorealistic relighting and rendering.
- Work with highly compressed 3D representations that maintain visual fidelity, enabled by the Sparse Compression VAE for efficient storage and sharing.
- Build upon an open-source research foundation designed with Responsible AI considerations, using ethically reviewed public datasets.
Pros & Cons
Pros
- Open-source
- High fidelity assets
- Handles complex structures
- Models surface attributes
- Supports PBR
- Photorealistic relighting
- Efficient pre-post processing
- O-Voxel encoding
- Sparse Compression VAE
- Minimal perceptual degradation
- Large-scale generative modeling
- Encodes geometry and appearance
- Handles arbitrary topologies
- Supports full PBR attributes
- Efficient Compression
- Instant image-to-3D conversion
- Flexible Dual Grids representation
- Compact structured latents
- Overcomes iso-surface field limitations
- 1536 PBR textured assets
- Supports Non-manifold geometry
- Handles enclosed interior structures
- Quick data conversions
- Free from rendering optimization
- Handles open surfaces
- 4B-parameter model
- Encapsulates fully textured 3D assets
- Voxel data compression
- Outperforms vanilla DiTs
- Handles Base Color, Roughness, Metallic, Opacity
- Research-focused tool
- Public datasets use
- Minimalistic 3D asset processing
Cons
- Purely a research project
- Requires high-end GPU
- Limited application scope
- Complex to implement
- Potential data biases
- No commercial use intended
- Not user-friendly
- Requires extensive technical knowledge
- Dependent on NVIDIA H100
- Absolute reliance on O-Voxel
Reviews
Rate this tool
Loading reviews...
❓ Frequently Asked Questions
TRELLIS.2 is an open-source image-to-3D model generator, designed to produce high fidelity textured assets using native 3D VAEs. Its core feature makes use of native and compact structured latents for providing both high fidelity and compression capabilities. This tool can handle complex structures, including open surfaces, non-manifold geometry, and enclosed interior structures, thus overcoming the limitations of iso-surface fields. It's a research project with Responsible AI considerations factored into all stages of its development.
TRELLIS.2 generates 3D models from images through a process that begins with an Instant Bidirectional Conversion that transforms meshes into a new representation termed O-Voxel. The Sparse Compression VAE then encodes these voxels into a compact Structured Latent space. The result is a highly compact representation of a fully textured 3D asset with negligible perceptual degradation.
Key features of TRELLIS.2 include the ability to handle complex structures, model arbitrary surface attributes such as base colour, roughness, metallic, and opacity, and optimize pre and post-processing of data for training and inference. Additional features include the utilization of 'O-Voxel', a novel 'field-free' sparse voxel structure to encode detailed geometry and complex appearance simultaneously, and a Sparse Compression VAE component for efficient voxel data compression.
TRELLIS.2 can handle complex structures including open surfaces, non-manifold geometry, and enclosed interior structures.
TRELLIS.2 handles non-manifold geometry by using a flexible dual grids representation within the O-Voxel structure. This methodology allows it to handle arbitrary topologies while preserving sharp edges.
TRELLIS.2 can model arbitrary surface attributes such as Base Color, Roughness, Metallic, and Opacity.
TRELLIS.2 facilitates Physically Based Rendering (PBR) by allowing for the modeling of arbitrary surface attributes such as Base Color, Roughness, Metallic, and Opacity. It can hence, accurately model rich surface materials for PBR and carry out photorealistic relighting.
The purpose of the 'O-Voxel' in TRELLIS.2 is to provide a novel 'field-free' sparse voxel structure designed to encode both precise geometry and complex appearance simultaneously.
TRELLIS.2 compresses voxel data using a Sparse Compression 3D VAE, employing a Sparse Residual Autoencoding scheme to directly compress voxel data into a compact structured latent space.
The benefits of using the Sparse Compression VAE component in TRELLIS.2 are twofold: It efficiently and compactly encodes the fully textured 3D asset with minimal perceptual degradation and it enables efficient large-scale generative modeling.
TRELLIS.2 is highly efficient in large-scale generative modeling. It can encapsulate fully textured 3D assets into a compact representation with minimal perceptual degradation, thus enabling it to carry out efficient large-scale generative modeling.
That TRELLIS.2 is an open-source project means it is freely available for the public to use, adapt, and improve upon. Any enhancements made to the source code can be shared back with the community, offering the opportunity for collaborative improvement.
In TRELLIS.2, Responsible AI considerations are factored into all stages of its development to ensure ethical use of AI. This includes using public datasets that have been reviewed to ensure there is no personally identifiable information or harmful content.
High fidelity textured assets generation in TRELLIS.2 refers to the creation of detailed, high-quality 3D assets from images, characterised by a high level of detail and realistic texture.
'Native and compact structured latents' in the context of TRELLIS.2 refers to the latent variables in its 3D VAE (Variational Autoencoder) model that encapsulate important, compact, and structured information about the input images for generating the 3D models.
TRELLIS.2 is considered a research project because its development and applications are purely for the purpose of research. It explores cutting-edge technologies in 3D generation methodologies, and its developments contribute to the wider scientific and AI research community.
In TRELLIS.2, the process of pre and post-processing of data involves simple methods for training and inference that enable instant conversions. These conversions are fully rendering-free and optimization-free, leading to efficient conversions, and in turn, high-quality 3D model generation.
Image-to-3D model generation as per TRELLIS.2 refers to the creation of a fully textured 3D asset from a two-dimensional image. It utilizes a number of advanced AI methodologies including native and compact structured latents, Sparse Compression VAE, and O-Voxel technology to generate these 3D models.
TRELLIS.2 overcomes the limitations of iso-surface fields by handling complex structures, including open surfaces, non-manifold geometry, and enclosed interior structures. It achieves this through the use of its O-Voxel technology which is designed to encode both detailed geometry and complex appearance simultaneously.
'3D VAEs' in the context of TRELLIS.2 refer to 3D Variational Autoencoders. These are a type of generative model used in TRELLIS.2 to produce high fidelity textured assets by learning and encoding the structure of the input data into a set of latent variables.
TRELLIS.2 facilitates Physically Based Rendering (PBR) by allowing for the modeling of arbitrary surface attributes such as Base Color, Roughness, Metallic, and Opacity. It can hence, accurately model rich surface materials for PBR and carry out photorealistic relighting.
The purpose of the 'O-Voxel' in TRELLIS.2 is to provide a novel 'field-free' sparse voxel structure designed to encode both precise geometry and complex appearance simultaneously.
TRELLIS.2 compresses voxel data using a Sparse Compression 3D VAE, employing a Sparse Residual Autoencoding scheme to directly compress voxel data into a compact structured latent space.
The benefits of using the Sparse Compression VAE component in TRELLIS.2 are twofold: It efficiently and compactly encodes the fully textured 3D asset with minimal perceptual degradation and it enables efficient large-scale generative modeling.
TRELLIS.2 is highly efficient in large-scale generative modeling. It can encapsulate fully textured 3D assets into a compact representation with minimal perceptual degradation, thus enabling it to carry out efficient large-scale generative modeling.
That TRELLIS.2 is an open-source project means it is freely available for the public to use, adapt, and improve upon. Any enhancements made to the source code can be shared back with the community, offering the opportunity for collaborative improvement.
In TRELLIS.2, Responsible AI considerations are factored into all stages of its development to ensure ethical use of AI. This includes using public datasets that have been reviewed to ensure there is no personally identifiable information or harmful content.
High fidelity textured assets generation in TRELLIS.2 refers to the creation of detailed, high-quality 3D assets from images, characterised by a high level of detail and realistic texture.
'Native and compact structured latents' in the context of TRELLIS.2 refers to the latent variables in its 3D VAE (Variational Autoencoder) model that encapsulate important, compact, and structured information about the input images for generating the 3D models.
TRELLIS.2 is considered a research project because its development and applications are purely for the purpose of research. It explores cutting-edge technologies in 3D generation methodologies, and its developments contribute to the wider scientific and AI research community.
In TRELLIS.2, the process of pre and post-processing of data involves simple methods for training and inference that enable instant conversions. These conversions are fully rendering-free and optimization-free, leading to efficient conversions, and in turn, high-quality 3D model generation.
Image-to-3D model generation as per TRELLIS.2 refers to the creation of a fully textured 3D asset from a two-dimensional image. It utilizes a number of advanced AI methodologies including native and compact structured latents, Sparse Compression VAE, and O-Voxel technology to generate these 3D models.
TRELLIS.2 overcomes the limitations of iso-surface fields by handling complex structures, including open surfaces, non-manifold geometry, and enclosed interior structures. It achieves this through the use of its O-Voxel technology which is designed to encode both detailed geometry and complex appearance simultaneously.
'3D VAEs' in the context of TRELLIS.2 refer to 3D Variational Autoencoders. These are a type of generative model used in TRELLIS.2 to produce high fidelity textured assets by learning and encoding the structure of the input data into a set of latent variables.
Pricing
Pricing model
Pricing
Paid options from
N/A
Related Videos
TRELLIS.2 (Microsoft), An Open-source Image to 3D generation Model
류젠에이아이(LiuGenAI)•110 views•Dec 18, 2025
Trellis: Versatile 3D Generation via Structured Latents
Gradient Daily•52 views•Nov 21, 2025
Image to 3D model AI - Microsoft TRELLIS | 100% Free
Pivot Point•4.6K views•Jan 24, 2025
Trellis2 震撼开源：40亿参数挑战商业级 3D 生成，PBR材质/透明纹理完美还原！🔥 本地部署保姆级教程
AI 技术分享•13 views•Dec 19, 2025
2025 3D Generation AI: Trellis, Hunyuan3D-2, TrioSG, Hi3DGen, MV-Adapter, Roblox-Cube
RauGenAI•189 views•Apr 10, 2025
Image to 3D- Running Trellis Locally with PC GPU
Unreal Polymath •9.7K views•Jan 5, 2025
Abhishek Does AI: Can AI Turn 2D Car Images into Printable 3D Models? (Trellis vs Hunyuan 3D-2)
Abhishek Does Stuff•50 views•May 14, 2025
