Stability AI, a leading AI technology company, announced the release of its new 3D video rendering model called Stable Video 3D (SV3D). Launched on Monday, this AI model is capable of transforming 2D images into immersive 3D videos, marking a significant advancement in the field of artificial intelligence.
Unlike other popular video generators like Sora by OpenAI, Pika 1.0, and Runway AI, Stable Video 3D does not rely on text inputs. Instead, it focuses solely on image inputs, leveraging advanced algorithms to convert static images into dynamic 3D models.
SV3D’s primary goal is to transform 2D images into 3D models. The company has released its new AI model for both non-commercial and commercial use.
The announcement was made via Stability AI’s official account on social media platform X. In the announcement, Stability AI highlighted SV3D’s innovative approach, describing it as a generative model based on Stable Video Diffusion. The company emphasized SV3D’s ability to deliver multi-view capabilities and superior quality compared to previous models.
Announcing Stable Diffusion 3, our most capable text-to-image model, utilizing a diffusion transformer architecture for greatly improved performance in multi-subject prompts, image quality, and spelling abilities.
Today, we are opening the waitlist for early preview. This phase… pic.twitter.com/FRn4ofC57s
— Stability AI (@StabilityAI) February 22, 2024
Stable Video 3D is offered in two variants: SV3D_p and SV3D_u. While SV3D_u generates orbital videos from single image inputs without camera conditioning, SV3D_p offers more versatility, supporting both orbital views and single images. This enables the creation of fully rendered 3D videos with specified camera paths.
According to the company, the newly launched AI model addresses the issues encountered by previous generations of models like Stable Zero123. SV3D utilizes mesh representations and Neural Radiance Fields (NeRF) to enhance the consistency and quality of rendered videos.
Stability AI, in a blog post, stated that Stable Video 3D addresses concerns regarding baked-in lighting, and integrates a disentangled illumination model. This model is optimized with the 3D texture and shape, aiming to enhance overall video consistency and quality. As a result, the AI model will be able to produce more realistic renderings.
Stability AI has made Stable Video 3D available for both non-commercial and commercial use. Commercial users can access the model through a Stability AI membership, for $20 per month for the Professional tier. As for non-commercial users, they have the option to download the model weights from Hugging Face, a popular platform for sharing AI models.
With this new AI model, SV3D is set to offer solutions to problems faced by older AI technologies used for video rendering. By using advanced techniques like Neural Radiance Fields and disentangled illumination models, SV3D aims to improve both the quality and consistency of rendered videos.
