Stability AI brings a new dimension to video with Stable Video 3D

Key Points:


  • Stable Video 3D (SV3D) released by Stability AI for rendering 3D video
  • SV3D provides new depth to video generation model with multi-view 3D mesh creation
  • SV3D can generate consistent and coherent novel multi-view images for 3D meshes



Stability AI has launched a new generative AI model called Stable Video 3D (SV3D), expanding its portfolio of video tools for rendering 3D videos. SV3D builds on the company’s existing video capabilities by enabling users to generate multi-view 3D meshes from a single input image. This new model is available for commercial use through a Stability AI Professional Membership or for non-commercial purposes via download from Hugging Face.


The company showcased the capabilities of SV3D in a video, demonstrating its ability to generate consistent and solid objects even as the camera rotates around them. According to Stability AI, SV3D is particularly valuable for creating immersive experiences in gaming and e-commerce.


Stability AI, known for its Stable Diffusion text-to-image models, including SDXL and Stable Diffusion 3.0, has previously introduced other 3D models like Stable Zero123. SV3D differentiates itself by generating multiple novel views simultaneously, improving the quality of 3D meshes from a single image compared to previous offerings.


In a research paper, Stability AI researchers detailed the techniques used to enable 3D generation from a single image using latent video diffusion. SV3D excels in generating coherent and consistent novel multi-view images, providing better quality and improved 3D meshes directly from the novel views it produces.


The SV3D model includes two versions: SV3D_u, which generates orbital videos based on single image inputs without camera conditioning, and SV3D_p, which accommodates both single images and orbital views, enabling users to create 3D videos along specified camera paths.



Prompt Engineering Guides



©2024 The Horizon