Meta raises the bar a notch above SLAB or Midjourney with an AI generating video from text.
For now, in the simple state of research conducted by machine-learning engineers within Facebook’s parent company, this new project bears the name, as simple as it is evocative, of ” Make-A-Video “.
IA, make me a video of a sheep drawing a sheep
While artificial intelligence generating an image from a text stirs up the crowds and hits the headlines within online art communities, Meta intends to offer video generation via the same process.
In the same vein as DALL-E or Midjourney, Meta’s artificial intelligence will generate a short five-second video similar to a GIF, based on a description given by the user. The examples given by the company in a dedicated article, quoted in source below, leave something to be desired for the moment, however.
The videos created are indeed still quite blurry and the animations lack naturalness. But it is after all a result to be expected from a young artificial intelligence which must feed on a lot of information (in this case images and videos) to develop.
A project still inaccessible to the public
For now, Make-A-Video is still at an early stage. Meta therefore does not currently allow everyone to test their artificial intelligence and is content to provide its own results. However, it is possible to register in order to be kept informed of the progress of the project and to hope to access a demo in the future.
Note that Meta is not the first to embark on the ambitious undertaking of an artificial intelligence generating a video from a text. Researchers from Tsinghua University and the Beijing Academy have indeed already started work at the beginning of the year on a project called ” CogVideo “.
Still, this is a rather impressive advance in the development of artificial intelligence. Meta argues that Make-A-Video advertises itself as ” an invaluable tool for designers and artists “.
Perhaps we will indeed be able to obtain copyright thanks to such works, but many ethical and practical questions arise on these exponentially successful processes. Questions to which humanity is still seeking answers in the face of the machine.
Source : Meta
3