Stable Point Aware 3D (SPA3D)
Stable Point Aware 3D (SPA3D) is an advanced feed-forward 3D reconstruction model developed by Stability AI that generates high-quality textured 3D meshes from a single input image in seconds.
Key Highlights
3D Model from Single Image
Creates detailed and textured 3D mesh models from a single 2D image using point-aware processing.
Texture Quality with UV Mapping
Provides realistic results by applying high-quality textures to 3D model surfaces with automatic UV mapping.
Fast 3D Generation
Enables rapid prototyping with 3D model generation within seconds thanks to the optimized pipeline.
Point-Aware Processing
Special algorithm that captures geometric details more accurately by considering point distribution in 3D space.
About
SPA3D (Single-view Point-cloud Aligned 3D) is an advanced AI model developed by Stability AI that performs high-quality 3D reconstruction from a single image. Unlike other single-view 3D reconstruction methods, it significantly improves geometric consistency by using point cloud alignment techniques. Integrated into the Stable Diffusion ecosystem, SPA3D is part of Stability AI's strategic investment in the 3D generation domain.
SPA3D's technical architecture is built on three core components. First is a diffusion-based view synthesis module that generates multi-view predictions from the input image. Second is the point cloud alignment system that extracts 3D point clouds from the generated views and uses these point clouds as geometric references. Third is a reconstruction module that produces high-quality mesh and texture from the point cloud reference. This feed-forward architecture produces results much faster than optimization-based methods. The model can generate 3D output from a single image in approximately 10 seconds on an A100 GPU.
In terms of performance, SPA3D presents impressive metrics. The approximately 10-second generation time on an A100 GPU provides a significant speed advantage compared to Zero123++'s 45-second duration. An F-Score of 0.452 on the GSO dataset represents a notable improvement when compared to One-2-3-45's score of 0.311. The point cloud alignment technique provides clear advantages particularly in preserving fine geometric details and accurately constructing complex structures.
SPA3D finds applications in rapid 3D prototyping, e-commerce product visualization, game asset generation, architectural pre-visualization, and augmented reality applications. It serves as a valuable tool particularly in e-commerce scenarios requiring quick 3D model creation from a single product photograph. Designers and 3D artists can use SPA3D to create reference models during prototyping stages.
SPA3D is made accessible as part of Stability AI's open-source strategy. Model weights and inference code are available through Hugging Face. Designed for compatibility with the Stable Diffusion ecosystem, it provides easy integration with existing Stability AI tools. Its PyTorch-based infrastructure is optimized for NVIDIA GPUs.
SPA3D is a significant work demonstrating the effectiveness of the point cloud alignment approach in single-image 3D reconstruction. Compared to Wonder3D's cross-domain attention approach and TRELLIS's structured latent representations, SPA3D offers an explicit and interpretable geometric reference system. Its integration with Stability AI's Stable Diffusion ecosystem makes SPA3D accessible to a broad developer community. The balance between speed and quality makes it a preferred solution particularly in production environments.
Delving deeper into SPA3D's technical approach, the impact of the point cloud alignment technique on geometric accuracy becomes better understood. Traditional single-view 3D reconstruction methods can encounter geometric inconsistencies when directly converting multi-view predictions to mesh. SPA3D addresses this problem by extracting a point cloud as an intermediate step and using this point cloud as a reference. The point cloud serves as an explicit representation of the object's structure in 3D space and guides mesh generation. The advantage of this approach is that errors are detected and corrected at an early stage. The model's feed-forward architecture avoids optimization loops, enabling fast inference and making it suitable for production environments. Its integration with Stability AI's Stable Diffusion ecosystem facilitates using SPA3D alongside existing image generation tools. For example, an image generated with Stable Diffusion can be directly fed as input to SPA3D and converted to a 3D model within seconds. This integrated workflow creates a powerful pipeline for creative processes and rapid prototyping scenarios.
Use Cases
Rapid 3D Prototyping
Quickly creating 3D prototypes from 2D drawings or photographs during the design process.
Game and AR/VR Asset Creation
Rapid 3D asset creation for game, augmented reality, and virtual reality projects.
E-Commerce 3D Visualization
Creating 3D models from product photos to offer interactive product experience.
Digital Twin Creation
Rapidly creating digital 3D copies of real-world objects for simulation and analysis.
Pros & Cons
Pros
- Stability AI's image-to-3D model — part of Stable 3D series
- Enhanced geometry with point cloud-based 3D awareness
- Practical use with fast inference time
- Published as open source
Cons
- Early-stage model — behind mature competitors
- Limited texture and material quality
- Uncertain future due to Stability AI's financial issues
- Inconsistencies in complex geometries
Technical Details
Parameters
Unknown
Architecture
Feed-forward 3D reconstruction
Training Data
Objaverse + proprietary
License
Stability AI Community License
Features
- Single image input
- Textured mesh
- UV mapping
- Fast generation
- GLB export
- Point-aware processing
- Multi-view generation
Benchmark Results
| Metric | Value | Compared To | Source |
|---|---|---|---|
| Üretim Süresi (Single Image) | ~10 saniye (A100) | Zero123++: ~45 saniye | Stability AI Blog |
| F-Score (GSO Dataset) | 0.452 | TripoSR: 0.421 | SPA3D Technical Report |
| Novel View PSNR | 21.5 dB | One-2-3-45: 19.2 dB | Papers With Code |
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.
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. The name reflects the core process: from one image, generate two to three to four to five views, then reconstruct a complete 3D object. In the first stage, a fine-tuned Zero123 model generates multiple novel views of the object from different angles based on the single input photograph. In the second stage, these generated multi-view images are fed into a cost-volume-based sparse-view reconstruction network that produces a textured 3D mesh with consistent geometry. Released in June 2023 under the MIT license, One-2-3-45 was among the first systems to demonstrate that combining 2D diffusion models with 3D reconstruction could produce reasonable 3D assets in under a minute. The model handles a variety of object types including everyday items, animals, vehicles, and artistic objects. Unlike optimization-based approaches like DreamFusion that require per-object optimization taking tens of minutes, One-2-3-45 runs in a feed-forward manner making it significantly faster. The output meshes include color and texture information and can be exported for use in standard 3D applications. As a fully open-source project with code available on GitHub, it has served as an influential reference for subsequent research in single-image 3D generation. The system is particularly useful for researchers and developers exploring rapid 3D content creation from limited input data.
Quick Info
Feed-forward 3D reconstruction