0
0
Stability AI Introduces Stable Video 3D Model
Stability AI expands its collection of generative AI models with the launch of Stable Video 3D (SV3D).
The newest addition to Stability AI’s repertoire, SV3D, is a generative AI video tool specifically tailored for rendering 3D videos. Drawing on the foundations of its Stable Video technology, which enables the creation of short videos from image or text prompts, SV3D represents a significant advancement by building upon the previous Stable Video Diffusion model, now optimized for the novel view synthesis and 3D generation tasks.
One of the key features of SV3D is its enhanced depth perception capability, allowing users to generate and manipulate multi-view 3D meshes from a single input image.
SV3D is now accessible for commercial use through the Stability AI Professional Membership, priced at $20 per month for creators and developers earning less than $1 million in annual revenue. For non-commercial applications, users have the option to download the model weights from Hugging Face.
The AI Impact Tour – Atlanta Stop
As part of the ongoing tour, Atlanta is the next destination for the AI Impact Tour, scheduled for April 10th. This exclusive event, held in collaboration with Microsoft, is invitation-only and will focus on discussions surrounding the transformative role of generative AI in the security workforce. Limited spots are available, so interested participants are encouraged to request an invite promptly.
An example video showcasing the capabilities of SV3D quickly generated reveals remarkable coherence and solidity of object forms even amidst slight distortions as the camera rotates.
Potential Applications in Game Creation and E-Commerce:
Stability AI has identified game creation and e-commerce as primary target sectors for SV3D application. By introducing camera path conditioning to their Stable Video Diffusion model, SV3D can effectively produce multi-view videos of objects, making it an invaluable resource for 3D asset generation in game development. Furthermore, SV3D enables the creation of 360-degree orbital videos that enhance the interactive and immersive shopping experience in e-commerce settings.
Evolution from Stable Zero123 to SV3D:
Known for its Stable Diffusion text-to-image generative AI models, Stability AI has a history of innovation starting with SDXL and leading up to the emerging Stable Diffusion 3.0. Notably, Stable Diffusion 1.5, an open-source image generation model, serves as the foundation for various AI image and video products, including Runway and Leonardo AI.
Transitioning from the previously released Stable Zero123 model, introduced in December 2023 for creating 3D images, SV3D represents a departure from its predecessor. SV3D pioneers a novel view synthesis approach to 3D generation, offering enhanced quality and efficiency in generating multiple unique views simultaneously from a single input image.
Advanced 3D Generation and Consistent Output:
Researchers at Stability AI outlined SV3D’s capabilities in a recently published research paper. The model’s ability to generate consistent multi-view images with seamless views from any perspective sets it apart from existing approaches, offering remarkable coherence and generalization in output quality.
In addition to its novel view synthesis functionalities, SV3D excels in optimizing 3D meshes by leveraging multi-view consistency to enhance the quality of 3D mesh representations directly from the generated novel views.
SV3D Variants for Diverse Applications:
SV3D introduces two powerful variants – SV3D_u and SV3D_p – tailored for specific use cases. SV3D_u specializes in creating orbital videos based on single image inputs without reliance on camera conditioning. In contrast, SV3D_p extends the capabilities of SV3D_u by accommodating both single images and orbital views, facilitating the creation of 3D videos along predefined camera paths.
Image/Photo credit: source url
