Stability AI Stable Video 3D

What is Stability AI Stable Video 3D?

Stability AI Stable Video 3D is a generative model that produces high-quality videos showing an object or scene from multiple viewpoints. It outperforms related models such as Stable Zero123 and Zero123-XL.

The model comes in two variants:

• SV3D_u: Creates orbital videos — a camera smoothly moving around an object — from a single image, with no need to specify a camera path.
• SV3D_p: Offers more control, generating 3D videos that follow specific camera paths you define, starting from a single image and using orbital views.

Licensing is straightforward: commercial use requires a Stability AI Membership, while the model weights can be downloaded directly for non-commercial or experimental use.

Under the hood, Stable Video 3D builds on Stable Video Diffusion. It optimizes 3D Neural Radiance Fields and mesh representations, improving the quality of 3D meshes generated from new viewpoints — especially in areas not fully visible in the initial predicted views.

Who created Stability AI Stable Video 3D?

Stable Video 3D was created by Stability AI, a company focused on developing advanced generative technology with a particular emphasis on 3D generation. The model improves both the quality of generated 3D content and the consistency of its views through techniques including masked score distillation sampling loss and a disentangled illumination model, which produces more realistic and varied lighting.

Key capabilities

• Generates multi-viewpoint videos from a single input image
• Supports both automatic orbital paths (SV3D_u) and user-defined camera paths (SV3D_p)
• Optimizes 3D Neural Radiance Fields and mesh representations for higher-quality 3D output
• Improves view consistency and lighting realism through novel training techniques

Availability and licensing

• Commercial use: Available with a Stability AI Membership
• Non-commercial use: Model weights are available for direct download

Stable Video 3D is a solid option for anyone working in 3D generation who needs reliable multi-viewpoint video synthesis from a single image, whether for experimentation or commercial production.