Stability AI Launches Open-Source Video Model

  • Stability AI addresses ethical concerns during the 'research preview' of Stable Video Diffusion.
  • What steps is Stability AI taking to overcome limitations in SVD and SVD-XT models?
Stability AI Unveils Stable Video Diffusion

This week Stability AI unveils Stable Video Diffusion, an innovative AI model designed to breathe life into static images, transforming them into dynamic videos. This move propels Stability into a select group, as one of the rare providers of open-source AI video-generating models available both for public and commercial use.

Stable Video Diffusion is Currently Undergoing

Nevertheless, Stability introduces a caveat — Stable Video Diffusion is currently undergoing what the company terms a "research preview." To harness the model's capabilities, users must adhere to specified terms of use, delineating intended applications such as educational or creative tools and design processes. However, the outlined terms also explicitly restrict the creation of factual representations of people or events.

History suggests that AI research previews, including Stability's own, may eventually find their way onto the dark web. This raises concerns about the potential misuse of Stability AI video, especially considering its apparent lack of a built-in content filter.

Stable Video Diffusion introduces two models: SVD and SVD-XT. SVD crafts 576x1024 videos in 14 frames from static images, while SVD-XT, sharing the same architecture, ups the frame count to 24. Both models exhibit flexibility in video-generating AI from three to 30 frames per second.

A whitepaper accompanying the release discloses that SVD and SVD-XT underwent initial training on a vast dataset of millions of videos. The subsequent "fine-tuning" phase involved a smaller set of hundreds of thousands to around a million clips. The origin of these videos remains unclear, potentially posing legal and ethical challenges if copyrighted material was inadvertently included.

The Models — both SVD and SVD-XT — Generate Fairly High-Quality Four-Second Clips

Despite potential concerns, both SVD and SVD-XT produce high-quality four-second clips, rivaling outputs from industry giants like Meta, and Google, and emerging startups Runway and Pika Labs. However, Stability acknowledges limitations, including the inability to generate videos without motion, respond to text prompts, render legible text, or consistently depict faces and people accurately.

Looking ahead, Stability envisions diverse applications for Stable Video Diffusion, planning to introduce additional models that build on and extend SVD and SVD-XT.

Stability AI faces challenges on the financial front. Reports from April indicate financial strain, prompting an executive hunt to bolster sales. Delays in wage payments and payroll taxes, coupled with a threat from AWS to revoke access to GPU instances, underscore the company's financial struggles.

Do you want to learn more about technology and AI? So, discover the news on Atlasiko!

Tetiana Rafalovych
Tetiana Rafalovych
Professional author in IT Industry

Author of captivating articles and news for Atlasiko Inc. I consistently deliver engaging content that captivates readers and keeps them coming back for more. I try to ensure that every piece is well-researched and informative. Whether it's news, in-depth features, or insightful analysis, I have a knack for transforming complex information into narratives that resonate with audiences.

Share your thoughts in the comments below!

Have any ideas or suggestions about the article or website? Feel free to write it.

Any Questions?

Get in touch with us by simply filling up the form to start our fruitful cooperation right now.

Please check your email
Get a Free Estimate