Stability AI has released a new generative AI model called Stable Video 4D. This model can transform a short video into eight new videos, each viewing the scene from a different angle.

Currently, the model takes a few-second video focused on a single object, such as a fish, and generates eight videos of it from eight different angles.

The model is available for experimentation and evaluation for research and development purposes on the HuggingFace platform. The company has not announced any plans for a commercial release.

The model features a new attention mechanism design. When generating each frame, it considers all neighboring frames from all angles, making the generated frames more coherent and consistent in terms of shape and smooth movement.

Although its name is 4D, it generates dynamic 3D videos. The name distinguishes it from the previous model, Stable Video 3D, which transforms images or text into 3D videos. The 4D model differs by generating dynamic videos from different angles, adding the camera’s position to its consideration, not just the 3D shape and appearance.

This model can be used in filmmaking to add visual effects. It can also be used by game developers to display objects from various angles and can be beneficial in creating virtual worlds.

The company is working to improve the model’s efficiency to generate longer videos with more complex scenes containing multiple objects.