A new product has been released by the emerging artificial intelligence company Stability AI called “Stable Video Spread”, which allows users to create a video from a single image.
The company wrote: “The intelligent and stable video generation model is based on the intelligent and stable image generation model, and this new model is an important step in our journey towards creating models for everyone of all kinds.”
The Stable Video Diffusion service is currently being provided as a research experiment by the company. Those who wish to use this model must agree to specific terms of use that define the intended uses for Stable Video Diffusion. This includes educational and creative applications, but not realistic representation applications of people or events.
Stability AI has clarified that interested users can register for the waiting list to access the distinguished future web experience that turns text into video. The tool offers potential applications in areas such as advertising, education, entertainment, and more. The Stable Video Diffusion product is currently only provided for research purposes and not for practical or commercial applications.
Stability AI has released a new tool that allows the conversion of images into videos, with the tool available in two models, each capable of creating between 14 and 25 frames. This tool operates at various speeds ranging from 3 to 30 frames per second, with a resolution of 576 x 1024 pixels.
The Stable Video Diffusion product can reassemble multiple displays from a single frame accurately, with effective dataset management.
The company stated: “We have found that these models outperform significant closed models in user preference studies when compared to text-to-video conversion platforms from Runway and Pika Labs.”
The tool produces relatively high-quality videos that are compatible with competing generation systems, but with certain limitations. The product generates short videos of 4 seconds duration, lacks perfect realism, cannot move the camera, lacks control over the text or the creation of readable text, and may not generate faces and people correctly.
The Stability AI tool was trained on gathering millions of video clips and later fine-tuned using a small dataset consisting of hundreds of thousands of video clips. The company announced that they use publicly available video clips for research purposes.
Videos are a primary target for generative artificial intelligence, as it facilitates content production, despite the high possibility of misuse through deep manipulation and violation of intellectual property rights.
