Startup Runway, which developed the popular neural network for generating images from a text description Stable Diffusion, has introduced a new Gen-1 language model that can transform existing videos into new ones by changing the style and composition of the frame.
The source data is a picture or a text description. As an example, the authors showed a video where a stack of books becomes a night metropolis, and a swimmer model becomes a real swimmer.
It looks like this:
So far, it has not been reported when the finished version will appear, whether it will be free like Stable Diffusion or not.
The authors themselves believe that the new neural network will become as popular as Stable Diffusion in image generation. They also believe that in the near future AI will generate a significant part of the video content.
Earlier it was reported that Google introduced the Dreamix neural network, which creates and edits videos from text, images, and other videos.