Google showed off a demo of an artificial intelligence product that can create videos up to five seconds long. The tool is called Lumiere. This is still beta, and it is not yet known when Lumiere will be available to users.

Lumiere is able to create realistic videos based on text inputs. It is also possible to move images partially or completely. Furthermore, Lumiere can imitate the style of an image, such as drawing, and then create videos using it. It is also possible to edit videos using the software. In one example Displayed by Googlenot only changes the color, but also the model of the dress that the woman is wearing, just by providing a text input.

In that paper On arXiv After it’s published, the Google research team describes how the program works. The team developed a new architecture called “Space-Time U-Net”. This makes it possible to create a video in one go. This should distinguish the architecture from existing models, which generate distributed keyframes in a first step, after which intermediate frames are interpolated with superior temporal resolution. Temporal super resolution is an image processing technology used to improve the temporal resolution of video. The goal is to create intermediate frames from existing frames in the video, effectively increasing the video frame rate. That’s not the case with Lumiere, which creates images without that high resolution.

The generated output is currently limited to videos that are only five seconds long and at a resolution of 1024 x 1024 pixels. Google itself considers the resolution to be low, but it is unclear whether future versions of the system will support higher resolution. Lumiere is currently a research project and therefore not yet available to the general public. It is not known when or if this will happen.