Stability AI unveils Stable Video 3D: A groundbreaking leap in 3D generative modeling

Stability AI, the creator of the renowned deep learning model Stable Diffusion, has unveiled Stable Video 3D (SV3D), a generative model based on Stable Video Diffusion, which was introduced last November. This move marks a significant step forward in the realm of 3D technology, promising enhanced quality and view-consistency.

The latest release, SV3D, improves it with two versions: SV3D_u and SV3D_p. The former generates orbital videos from single image inputs without camera conditioning, while SV3D_p extends the functionality to include both single images and orbital views, facilitating the creation of 3D video along specified camera paths.

According to Stability AI, the adaptation of their image-to-video diffusion model with camera path conditioning enables Stable Video 3D to produce multi-view videos of an object. This results in significant improvements in generalization and view-consistency of the generated outputs.

Stable Video 3D represents a substantial leap in 3D generation, particularly in the area of novel view synthesis (NVS). The model is capable of delivering coherent views from any angle, enhancing pose-controllability and ensuring consistent object appearance across multiple views. These features contribute to the realism and accuracy of 3D generations.

Stable Video 3D is now available for commercial use with a Stability AI Membership. For non-commercial use, the model weights can be downloaded on Hugging Face.