Google has presented an innovative tool artificial intelligence (generative AI) known as Lumierecapable of transforming a simple description into text or an image into a video, in addition to allowing editing and working on specific parts of the image.
The highlight of Lumiere is its ability to generate realistic videos with coherent motion and edit them to achieve specific styles. This diffusion model developed by Google is capable of efficiently creating videos in a single step, determining the temporal duration of the video, all thanks to its architecture called 'Space-Time U-Net' (STUNet), which reduces the temporal signal and spatial, performing most of the computational work in a compact representation. The result is a 5-second video that can be generated quickly.
Lumiere's versatility extends to both generating videos from a text description and editing to achieve a particular style.
In addition, it is possible to create videos from an image if an accompanying text indication is included, as Google reports on the GitHub page dedicated to this project.
Lumiere's capabilities are amazing, including animating specific parts of an image and reconstructing missing segments in a video for a complete display. This new tool promises to revolutionize video creation and editing, offering users an efficient and fast way to bring their ideas and audiovisual projects to life.