One-2-3-45
One-2-3-45 is a single-image 3D reconstruction system developed by researchers at UC San Diego that generates textured 3D meshes from a single input image through a two-stage pipeline combining multi-view generation with sparse-view 3D reconstruction.
Key Highlights
Pioneering Multi-View-Then-Reconstruct Paradigm
One of the first systems to demonstrate the two-stage approach of generating multi-view images then performing 3D reconstruction, establishing a pattern now standard in the field
45-Second Complete Pipeline
Completes the entire single-image to textured 3D mesh pipeline in approximately 45 seconds, dramatically faster than optimization-based methods requiring hours
Cost-Volume 3D Reconstruction
Uses cost-volume-based reconstruction that aggregates information from multiple generated views, providing robustness against view inconsistencies
MIT Licensed Academic Research
Fully open-source research from UC San Diego under MIT license with reproducible code, serving as an important reference for the 3D generation research community
About
One-2-3-45 is a single-image 3D reconstruction system developed by researchers at UC San Diego that generates textured 3D meshes from a single input image through a two-stage pipeline combining multi-view generation with sparse-view 3D reconstruction. The name reflects the process: from one image, generate 2D multi-view images, then reconstruct 3D geometry, achieving this in approximately 45 seconds. The model left a lasting and profound impact on the field as one of the pioneering works demonstrating the viability of the multi-view-then-reconstruct paradigm.
The system operates in two key stages. First, a view-conditioned 2D diffusion model (based on Zero123) generates multiple consistent views of the object from different angles. Second, a cost-volume-based sparse-view reconstruction module processes these generated multi-view images to produce a 3D mesh with texture maps. This two-stage approach separates the challenging tasks of view synthesis and 3D reconstruction into manageable sub-problems. The ability to independently optimize each stage facilitates improvement of the system's overall performance and allows different components to be developed separately.
One-2-3-45's contribution lies in demonstrating that combining pre-trained 2D diffusion models with multi-view 3D reconstruction is a viable and efficient approach to single-image 3D generation. The system achieves reasonable 3D reconstruction quality in under a minute, which was a significant improvement over optimization-based methods at the time of publication that could take hours per object. This speed advantage made iterative design cycles practical in 3D content creation workflows and provided researchers with the ability to perform rapid experimental iteration.
The reconstruction module uses a cost-volume approach that aggregates information from multiple generated views to estimate 3D geometry. This approach is more robust to inconsistencies between generated views than methods that rely on single-view depth estimation. The cost volume evaluates different depth hypotheses to determine the most likely 3D geometry, coherently combining information from multiple views throughout this process. The resulting meshes include both geometry and texture information, providing usable 3D assets for visualization and prototyping purposes.
Among the system's limitations, inconsistencies between generated views can affect the final 3D reconstruction quality. These inconsistencies may become more pronounced for objects with complex geometries or fine details. However, the robustness of the cost-volume approach significantly mitigates the impact of such inconsistencies, maintaining reasonable quality across a variety of input types and object categories.
Released under the MIT license, One-2-3-45 is fully open-source with code and pre-trained weights available on GitHub. While newer models like InstantMesh and TripoSR have since achieved higher quality and faster generation, One-2-3-45 remains historically important as one of the first systems to demonstrate the multi-view-then-reconstruct paradigm that has become standard in the field. The architectural design principles established by the model continue to form the foundation of subsequent research and applications in single-image 3D reconstruction.
Use Cases
3D Generation Research Baseline
Serves as a standard comparison baseline for evaluating new single-image 3D reconstruction methods in academic research publications
Quick 3D Prototyping
Generate rough 3D models from reference images in under a minute for design prototyping and concept visualization purposes
Educational Tool for 3D AI
Learn about multi-view 3D reconstruction concepts through a well-documented, accessible implementation with clear two-stage pipeline design
Pipeline Architecture Reference
Use as an architectural reference for building custom 3D generation pipelines that combine 2D diffusion with 3D reconstruction modules
Pros & Cons
Pros
- Creates 3D model from a single 2D image in 45 seconds
- Zero-shot approach — no retraining needed for each object
- Consistent angle synthesis with multi-view diffusion
- Open-source research project
Cons
- Mesh quality behind commercial tools
- Quality loss in fine details and edge areas
- Difficulty with asymmetric objects
- Limited texture quality
Technical Details
Parameters
N/A
License
MIT
Features
- Single Image to 3D
- Multi-View Generation Stage
- SparseView Reconstruction
- Zero123 Based Pipeline
- Open-Source MIT License
- UC San Diego Research
- Mesh Output with Textures
- Academic Reference Implementation
Benchmark Results
| Metric | Value | Compared To | Source |
|---|---|---|---|
| Novel View PSNR | 18.8 dB (GSO) | Unique3D: 20.1 dB | arXiv 2306.16928 |
| Üretim Süresi | ~45 saniye | Wonder3D: ~3 dakika | GitHub One-2-3-45 |
| SSIM (GSO) | 0.842 | Unique3D: 0.922 | arXiv 2306.16928 |
Available Platforms
Frequently Asked Questions
Related Models
Era3D
Era3D is a multi-view generation model developed by Alibaba that produces high-resolution, camera-aware multi-view images and normal maps from single input images for 3D reconstruction. The model introduces two key innovations that address common limitations in multi-view generation: a focal length estimation module that adapts to the camera perspective of the input image, and an efficient row-wise attention mechanism that enables generation at higher resolutions than competing methods while using less GPU memory. Era3D generates six consistent views along with corresponding normal maps at 512x512 resolution, providing rich geometric information for downstream 3D mesh reconstruction. The camera-aware design means the model can handle input images taken from different perspectives and focal lengths without degradation in output quality, a significant improvement over methods that assume a fixed camera model. The row-wise attention mechanism replaces the computationally expensive full cross-view attention with a more efficient alternative that processes attention along horizontal rows, reducing memory requirements while maintaining view consistency. Released in May 2024 under the Apache 2.0 license, Era3D is fully open source with code and pre-trained weights available on GitHub. The model demonstrates strong performance across diverse object categories and produces clean multi-view outputs suitable for high-quality 3D reconstruction. Era3D is particularly valuable for professional 3D content creation workflows where input images come from varied sources with different camera characteristics, and where high-resolution multi-view generation is essential for capturing fine details in the final 3D 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 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.