Generative AI startup Stability AI has expanded its AI capabilities with the launch of Stable Video Diffusion, a new AI tool that breathes life into static images by generating fluid video animations.
Building upon the foundation of their popular text-to-image model, Stable Diffusion, Stability AI is contributing to the democratisation of generative AI video models through the release of its Stable Video Diffusion.
The start-up announced that Stable Video Diffusion is its, “First foundation model for generative video based on the image model Stable Diffusion,” said the company in an announcement. “Now available in research preview, this state-of-the-art generative AI video model represents a significant step in our journey toward creating models for everyone of every type.”
During the research preview, Stability AI shared the code for Stable Video Diffusion on GitHub. Users interested in the model can get the necessary weights for local use from Stability AI’s Hugging Face page. The company also shared detailed technical information about the model in a dedicated research paper.
Bringing images to life
Stable Video Diffusion consists of two image-to-video models, generating 14 and 25 frames, as explained in the paper. These models aren’t very common because they allow users to adjust frame rates from 3 to 30 frames per second, offering flexibility in creative projects.
“At the time of release in their foundational form, through external evaluation, we have found these models surpass the leading closed models in user preference studies,” said the start-up.
Stability AI is enthusiastic about Stable Video Diffusion’s possibilities but clarifies that, for now, it’s not meant for real-world or commercial use. The company is actively seeking user feedback to improve the model’s safety and overall quality.
Interested users can sign up for the waitlist and access what the company describes as, “A new upcoming web experience featuring a Text-To-Video interface.”