Stable Video Diffusion 2

Generative model for high-quality 3D generation from single images with improved view consistency. Stable Video 3D (SV3D) is a revolutionary generative model that advances 3D technology. Based on Stable Video Diffusion, SV3D significantly improves the quality and view-consistency for novel view synthesis and 3D generation from single images. It offers superior pose-controllability and consistent object appearance across multiple views, producing accurate and realistic 3D meshes directly from single image inputs. SV3D is available for commercial purposes with a Stability AI Membership and offers open-source access for non-commercial use via Hugging Face.

Stability AI licenses offer flexible generative AI solutions with self-hosting benefits and access to models like SV3D. Stability AI licenses offer flexibility for your generative AI needs by combining our range of state-of-the-art open models with self-hosting benefits. SV3D takes a single object image as input and output novel multi-views of that object. We can then use those novel-views and SV3D to generate 3D meshes. SV3D comes in two variants tailored for different needs. SV3D_u focuses on generating orbital videos from single image inputs without camera conditioning, making it ideal for straightforward 3D visualizations. On the other hand, SV3D_p extends this capability to include both single images and orbital views, allowing for more complex 3D video creation along specified camera paths. SV3D is available for commercial use with a Stability AI Membership, offering a wide range of applications in various industries. For non-commercial purposes, model weights can be downloaded from the Hugging Face platform, accompanied by our comprehensive research paper.