Note: Stable Video 3D weights are released under a Stability AI Non-Commercial Research Community License, and cannot be utilized for commercial purposes. Please read the license to verify if your use case is permitted.
Lab Station OS
The simplest way to self-host Stable Video 3D. Launch a dedicated cloud GPU server running Lab Station OS to download and serve the model using any compatible app or framework.
Direct Download
Download model weights for local inference. Must be used with a compatible app, notebook, or codebase. May run slowly, or not work at all, depending on your system resources, particularly GPU(s) and available VRAM.
Generate images and videos using a powerful low-level workflow graph builder - the fastest, most flexible, and most advanced visual generation UI.
Model Report
stabilityai / Stable Video 3D
Stable Video 3D (SV3D) generates 360-degree orbital videos from single images. Two variants exist: SV3D_u for automatic camera paths and SV3D_p for user-defined orbits. Trained on Objaverse, it creates 21-frame sequences at 576x576 resolution by inferring 3D structure from 2D inputs.
Explore the Future of AI
Your server, your data, under your control
Stable Video 3D (SV3D) represents a significant advancement in generative AI technology, developed by Stability AI as an evolution of their Stable Video Diffusion framework. This innovative model specializes in creating orbital videos from single static images, offering new possibilities for 3D content generation. Full details are available in the technical report and demonstrated in their video summary.
Model Architecture and Variants
The SV3D family consists of two primary variants, each serving distinct use cases while maintaining the core capability of generating 21-frame videos at 576x576 resolution:
SV3D_u: The base variant focuses on unconditional video generation from single images without camera path conditioning. This model serves as the foundation for the family's capabilities.
SV3D_p: Building upon SV3D_u, this enhanced variant adds support for camera path conditioning. It can accept both single images and orbital view inputs, allowing for more controlled 3D video generation along specified camera trajectories.
Both variants are built upon Stability AI's existing Stable Video Diffusion technology, as detailed in their GitHub repository.
Training and Dataset
The model's training process utilized a carefully curated subset of the Objaverse dataset, which is available under the CC-BY license. Stability AI implemented an enhanced rendering methodology during training to improve the model's ability to generalize to real-world scenarios. This approach helps bridge the gap between synthetic training data and practical applications.
The training methodology focused on enabling the model to understand and reproduce 3D characteristics from 2D inputs, allowing it to generate convincing orbital movements around subjects in the generated videos. More technical details about the training process can be found in the project's arXiv paper.
Usage and Licensing
SV3D is released under the Stability AI Community License, with separate terms available for commercial usage. The model comes with specific usage guidelines and restrictions:
The model should not be used for generating factual representations of people or events
All usage must comply with Stability AI's Acceptable Use Policy
Commercial applications require separate licensing arrangements
Detailed implementation instructions and code examples are available through the official generative models repository. The project's dedicated website provides additional resources and demonstrations of the model's capabilities.
References
Technical Report - Detailed technical documentation of the model
Video Summary - Visual demonstration of model capabilities
