Stable Video 3D can create videos of rotating 3D objects or generate 3D video with a specific ‘camera’ path.
Stability AI is releasing two versions of a new generative AI model called Stable Video 3D (SV3D). Both versions can turn image prompts into 3D video “meshes” that can show an “orbiting” 360-degree view of an object, while the more advanced version can also incorporate “specified camera paths” into its video generation, according to the company.
A blog post from Stability AI shows how the model is designed to output 3D model videos of objects like fashion accessories and toy animals without being given images of the subject from all angles.
“This is mainly made possible by the large-scale pre-training of generative models which enables sufficient generalization to various domains,” Stability AI’s researchers write in their report.
