LGM
LGM (Large Gaussian Model) is a 3D generation model developed by researchers at Peking University that produces high-quality 3D objects from single images or text prompts in approximately five seconds using 3D Gaussian Splatting representation.
Key Highlights
5-Second Gaussian Splatting Generation
Generates complete 3D Gaussian Splatting representations from single images in approximately 5 seconds, balancing speed with high visual quality output
Real-Time Renderable Output
Produces 3D Gaussian representations that enable real-time rendering from any viewpoint without neural network inference, suitable for interactive 3D applications
Multi-View Consistency
Generates four consistent orthogonal views before reconstruction, ensuring accurate geometry from multiple perspectives for robust 3D shape recovery
Dual Output: Gaussians and Meshes
Supports both 3D Gaussian Splatting format for real-time rendering and mesh extraction pipeline for traditional 3D workflow compatibility
About
LGM (Large Gaussian Model) is a 3D generation model developed by researchers at Peking University that generates 3D objects from single images in approximately 5 seconds using 3D Gaussian Splatting representation. Released in 2024, LGM combines multi-view image generation with Gaussian-based 3D reconstruction to produce high-quality 3D assets that can be rendered in real-time using Gaussian splatting renderers. The model played a pioneering role in the field as one of the first works to successfully demonstrate the integration of Gaussian Splatting into 3D generation pipelines.
The model operates through a two-stage pipeline. First, a multi-view diffusion model generates four consistent orthogonal views of the object from the input image. These four views are then processed by the Large Gaussian Model, which predicts a set of 3D Gaussians that represent the object's geometry, appearance, and transparency. The resulting Gaussian representation can be rendered from any viewpoint in real-time using splatting-based renderers. The asymmetric U-Net architecture directly regresses Gaussian parameters from multi-view images, enabling fast inference without requiring additional optimization steps. Each individual Gaussian element encodes its position in 3D space, scale, orientation, and color information.
LGM's use of 3D Gaussian Splatting as the output representation offers several advantages. Gaussian splatting enables real-time rendering without the computational overhead of neural radiance fields, making the generated assets immediately usable in interactive 3D applications. The representation naturally handles view-dependent effects like specular highlights and translucency, providing photorealistic visualization. For applications requiring traditional mesh format, LGM includes a mesh extraction pipeline that converts the Gaussian representation to textured polygonal meshes. This dual output flexibility makes the model suitable for both interactive visualization and traditional 3D workflows.
The model's 5-second generation time represents an excellent balance between speed and quality. While not as fast as TripoSR (sub-second), LGM typically produces higher visual quality output, particularly in terms of view consistency and surface detail. The generation time is fast enough for interactive workflows while delivering results competitive with slower optimization-based methods. Parameters including per-Gaussian position, scale, rotation, opacity, and spherical harmonic coefficients are learned, and this rich parameter set enables detailed appearance modeling across diverse object types.
LGM was trained on the Objaverse dataset and processes input images at 512x512 resolution. The model performs particularly well on objects with smooth surfaces, while it may show limitations on objects with very fine geometric details or complex internal structures. The number and distribution of Gaussians in the output automatically adjust according to geometry complexity, with typically thousands of Gaussian elements used per object.
Released under the MIT license, LGM is fully open-source with code and pre-trained weights available on GitHub. The model has been influential in demonstrating the viability of Gaussian splatting as an output format for 3D generation models and has contributed to the growing ecosystem of Gaussian-based 3D tools and applications. The research community continues to build upon LGM's approach to develop higher-resolution and more detailed Gaussian-based 3D generation methods.
Use Cases
Interactive 3D Web Experiences
Generate Gaussian splatting assets for web-based 3D viewers that render in real-time, creating interactive product showcases and virtual galleries
Rapid 3D Asset Prototyping
Create 3D prototypes from concept images in seconds for design review, client presentations, and iterative creative development processes
Real-Time 3D Applications
Feed generated Gaussian assets into real-time applications including AR experiences, interactive demos, and spatial computing environments
3D Content Pipeline Integration
Integrate into automated content pipelines where images are converted to 3D assets at scale for catalogs, inventories, and digital twin creation
Pros & Cons
Pros
- Generates 3D objects from image or text within 5 seconds at 512x512 resolution with up to 65,536 Gaussians
- ECCV 2024 Oral paper — demonstrates superior visual quality compared to DreamGaussian and TriplaneGaussian
- Effectively addresses blurry back views and flat geometry common in prior image-to-3D methods
- Gaussian Splatting representation is more expressive and faster to render than triplane-based NeRFs
- Achieves high-resolution generation (512x512) while maintaining fast 5-second generation speed
Cons
- Output quality is inherently tied to upstream multi-view diffusion model quality — inconsistent inputs degrade results
- May not follow text prompts effectively for unconventional or unusual objects
- Limited to object-centric scenes — cannot handle full scene reconstruction
- Requires multi-view images as input, adding dependency on separate diffusion model for generation pipeline
- Gaussian Splatting output requires additional conversion for use in standard 3D applications
Technical Details
Parameters
N/A
License
MIT
Features
- Single Image to 3D Gaussian
- Ultra-Fast 5-Second Generation
- 3D Gaussian Splatting Output
- High-Quality Multi-View Synthesis
- Mesh Extraction Support
- Open-Source MIT License
- Peking University Research
- Real-Time 3D Rendering
Benchmark Results
| Metric | Value | Compared To | Source |
|---|---|---|---|
| Üretim Süresi | ~5 saniye | InstantMesh: ~10 saniye | ECCV 2024 / arXiv 2402.05054 |
| Eğitim Çözünürlüğü | 512×512 px | OpenLRM: 256×256 | GitHub 3DTopia/LGM |
| Gaussian Sayısı | ~40K 3D Gaussian | — | ECCV 2024 Paper |
| Novel View PSNR | 21.5 dB (GSO) | InstantMesh: 22.2 dB | arXiv 2402.05054 |
Available Platforms
Frequently Asked Questions
Related Models
InstantMesh
InstantMesh is a feed-forward 3D mesh generation model developed by Tencent that creates high-quality textured 3D meshes from single input images through a multi-view generation and sparse-view reconstruction pipeline. Released in April 2024 under the Apache 2.0 license, InstantMesh combines a multi-view diffusion model with a large reconstruction model to achieve both speed and quality in single-image 3D reconstruction. The pipeline first generates multiple consistent views of the input object using a fine-tuned multi-view diffusion model, then feeds these views into a transformer-based reconstruction network that predicts a triplane neural representation, which is finally converted to a textured mesh. This two-stage approach produces significantly higher quality results than single-stage methods while maintaining generation times of just a few seconds. InstantMesh supports both text-to-3D workflows when combined with an image generation model and direct image-to-3D conversion from photographs or artwork. The output meshes include detailed geometry and texture maps compatible with standard 3D software and game engines. The model handles a wide variety of object types including characters, vehicles, furniture, and organic shapes with good geometric fidelity. As an open-source project with code and weights available on GitHub and Hugging Face, InstantMesh has become a popular choice for developers building 3D asset generation pipelines. It is particularly useful for game development, e-commerce product visualization, and rapid prototyping scenarios where fast turnaround and reasonable quality are both important requirements.
Meshy
Meshy is a proprietary AI-powered 3D generation platform developed by Meshy AI that creates detailed, production-ready 3D models from text descriptions and images. The platform combines text-to-3D and image-to-3D capabilities with advanced AI texturing features, positioning itself as a comprehensive solution for rapid 3D content creation. Meshy uses a transformer-based architecture that generates textured 3D meshes with PBR-compatible materials, making outputs directly usable in game engines like Unity and Unreal Engine without additional processing. The platform offers multiple generation modes including text-to-3D for creating objects from written descriptions, image-to-3D for converting photographs into 3D models, and AI texturing for applying realistic materials to existing untextured meshes. Generated models include proper UV mapping, normal maps, and physically based rendering materials suitable for professional workflows. Meshy provides both a web-based interface and an API for programmatic access, making it accessible to individual artists and scalable for enterprise pipelines. The platform is particularly popular among game developers, animation studios, and AR/VR content creators who need to produce large volumes of 3D assets efficiently. As a proprietary commercial service launched in 2023, Meshy operates on a subscription model with free tier access for limited generations. The platform continuously updates its models to improve output quality, topology optimization, and texture fidelity, competing directly with other AI 3D generation services in the rapidly evolving market.
Meshy v4
Meshy-4 is a historical model version retired for API requests. Check the retirement boundary, migration dependencies and current plan terms before starting new work.
OpenLRM
OpenLRM is an open-source implementation of the Large Reconstruction Model architecture for single-image 3D reconstruction, developed by Zexiang Xu and collaborators. The project provides a fully open and reproducible implementation of the LRM approach, which uses a transformer-based architecture to predict 3D representations from single input images in a feed-forward manner. OpenLRM processes an input image through a pre-trained vision encoder like DINOv2, then feeds the resulting features into a transformer decoder that generates a triplane-based neural radiance field representation, which can be rendered from novel viewpoints or converted to a textured 3D mesh. The entire reconstruction takes only a few seconds on a modern GPU, making it practical for interactive applications and batch processing workflows. Released under the Apache 2.0 license in December 2023, OpenLRM fills a critical gap in the 3D AI research community by providing an accessible reference implementation that researchers can study, modify, and build upon. The model supports various output formats and can be integrated into existing 3D pipelines for applications ranging from game development to e-commerce product visualization. OpenLRM handles diverse object categories including furniture, vehicles, characters, and everyday items with reasonable geometric fidelity. Pre-trained model weights are available on Hugging Face for immediate use. As one of the foundational open-source projects in feed-forward 3D reconstruction, OpenLRM has directly influenced and enabled numerous downstream projects and research efforts in the rapidly evolving single-image 3D generation space.