Stable Virtual Camera

Stable Virtual Camera launched by Stability AI can convert ordinary 2D pictures into three-dimensional 3D videos. Users can easily generate smooth 3D effect videos by simply setting the camera movement trajectory (such as spiral, zoom, pan, etc.).

Core functions:

• 2D to 3D: Convert static images into dynamic 3D videos.

• Flexible camera control: Users can customize camera movement to achieve various visual effects.

• Multi-format output: Supports multiple video ratios to meet the needs of different platforms.

• High-quality generation: The generated 3D video has realistic effects and smooth graphics.

• Ultra-long video support: Up to 1000 frames of video can be generated.

Technical Highlights:

• Use generative AI and neural rendering technology.

• Optimize consistency and smoothness of multi-angle images.

• Based on the diffusion model, gradually improve the quality of video.

Project resources:

• Official website: https://stable-virtual-camera.github.io/

• GitHub code base: https://github.com/Stability-AI/stable-virtual-camera

• Hugging Face Model: https://huggingface.co/stabilityai/stable-virtual-camera

• arXiv paper: https://arxiv.org/pdf/2503.14489

Applicable scenarios:

• Advertising and product display.

• Creative content production.

• Education and training.

Installation steps

• Clone the repository:

   git clone --recursive https://github.com/Stability-AI/stable-virtual-camera
  cd stable-virtual-camera

• Installation dependencies:

   pip install -e .

• Require:

  Please note that Windows users need to run via WSL (Windows Subsystem for Linux) because native Windows does not support Flash.

• Python version: >=3.10

• PyTorch version: >=2.6.0

• Additional dependencies:

• If you want to use demo or develop, please refer to INSTALL.md for more dependencies.

How to use

• Hugging Face Authentication:

• You need to log in through Hugging Face to download the model weights.

• Run the following command to authenticate:

   huggingface-cli login

• Demo program:

• Gradio Demo: Suitable for ordinary users and operates through a graphical interface.

   python demo_gr.py

  For detailed guides, see GR_USAGE.md .

• CLI Demo: Suitable for advanced users or researchers, allowing for finer granular control.

   python demo.py --data_path <data_path> [additional arguments]

  For detailed guides, see CLI_USAGE.md .

• More information:

• If you want to benchmark NVS models, check out the relevant information in the paper for detailed data on scenarios, segmentation, and input/target views.