
Trellis
Microsoft Trellis is a state-of-the-art image-to-3D generation model that utilizes Structured Latent (SLAT) representations and transformer architectures to reconstruct high-fidelity 3D assets from single images. It generates production-ready outputs in multiple formats including textured meshes, 3D Gaussian splats, and radiance fields with exceptional geometric accuracy and surface details.
Overview
Trellis is a 3d generation model available on the GenVR platform. Microsoft Trellis is a state-of-the-art image-to-3D generation model that utilizes Structured Latent (SLAT) representations and transformer architectures to reconstruct high-fidelity 3D assets from single images. It generates production-ready outputs in multiple formats including textured meshes, 3D Gaussian splats, and radiance fields with exceptional geometric accuracy and surface details.
Key Features
- Structured Latent (SLAT) representation for efficient 3D encoding
- Multi-format output generation (meshes, Gaussian splats, radiance fields)
- Single-image 3D reconstruction with detailed geometry
- Transformer-based architecture for scalable generation
- High-resolution texture synthesis and UV mapping
- Fine-grained topology handling and complex surface reconstruction
- Support for both object-centric and scene-level generation
- Optimized inference pipeline for API deployment
Popular Use Cases
- Converting product photography into interactive 3D models for online retail configurators
- Generating game-ready props and environmental assets from concept art or reference images
- Creating 3D training data and synthetic datasets for computer vision model development
- Producing Gaussian splat assets for immersive web-based AR experiences and virtual showrooms
- Rapid prototyping of physical product designs from hand-sketched or photographed concepts
Best For
- Game developers needing rapid asset prototyping and environmental prop generation
- E-commerce platforms requiring automated 3D product visualization from catalog photos
- AR/VR creators building spatial experiences with photorealistic object integration
- 3D artists and designers accelerating concept modeling and look-development workflows
- Architectural visualization studios converting reference photos into 3D scene elements
Limitations to Keep in Mind
- Single-view ambiguity may cause geometric inaccuracies for highly occluded or symmetrical objects
- Limited support for transparent, reflective, or complex multi-material surfaces
- Requires moderate GPU compute resources for high-resolution generation
- May struggle with extreme novel-view extrapolation far from the input camera angle
- Generation quality depends heavily on input image resolution and lighting conditions
Why Choose This Model
- Superior Geometry Fidelity: Reconstructs intricate surface details and complex topologies that surpass traditional photogrammetry methods.
- Format Flexibility: Generates assets in meshes, Gaussian splats, or radiance fields to match specific pipeline requirements without reprocessing.
- Single-Image Input: Creates complete 3D models from just one photograph, eliminating the need for multi-view capture setups.
- Production-Ready Quality: Outputs textured, game-engine-compatible assets ready for immediate integration into Unity, Unreal, or web-based 3D viewers.
- Research-Backed Architecture: Built on Microsoft's cutting-edge SLAT technology ensuring state-of-the-art reconstruction accuracy and consistency.
- Rapid Generation Speed: Produces complete 3D assets in seconds through optimized transformer inference, accelerating creative workflows.
- Texture Preservation: Maintains high-fidelity surface colors and material properties from source images during the 3D conversion process.
- Occlusion Handling: Intelligently hallucinates plausible geometry for hidden regions based on learned 3D priors and semantic understanding.
- Scalable API Integration: Designed for robust cloud deployment with consistent performance across varying input complexities.
- Cross-Industry Versatility: Handles diverse object categories from products and furniture to characters and architectural elements.
- Reduced Manual Modeling: Cuts 3D asset creation time from hours to minutes, significantly lowering production costs.
- AR/VR Optimization: Generates lightweight Gaussian splat representations ideal for immersive spatial computing experiences.
Alternatives on GenVR
- Hyper 3D Rodin v2
- Hunyuan 2.1 - 3D
- Trellis 2
Pricing
Billed through GenVR credits
Properties
Customizable parameters available for this model.
Required
List of input images to generate 3D asset from
Optional
Random seed for generation
GLB Extraction - Texture Size (only used if generate_model=True)
Generate color video render
Generate 3D model file (GLB)
Randomize seed
GenVR Visual App
Experience the power of Trellis through our intuitive visual interface. Experiment with prompts, adjust parameters in real-time, and download your results instantly.
Launch AppDeveloper API Docs
Integrate this model into your own applications. Access enterprise-grade performance, scalable infrastructure, and detailed documentation for rapid deployment.
Explore APIMore in 3D Generation
Discover other high-performance models in the same category as Trellis.