Meta introduces Emu Video, artificial intelligence that can create video from text

Meta introduced Emu Video, an evolution of Emu generative artificial intelligence (AI) that can create videos based on text description. Tool can create short, low-resolution clips in a few seconds.

Emu Video is like an image generator. The tool interprets the user’s description, processes the request in its database, and returns a four-second video at 512 x 512 resolution at 16 frames per second.

“This is a unified architecture for video rendering tasks that can respond to a variety of inputs: text only, image only, and text and image. We divide the process into two steps: first, we create images conditioned by a text prompt, and then we create images conditioned based on both the text and the rendered image create a video,” explains Meta.

According to Meta, this “split” approach allows you to train generative models more efficiently. The results are more convincing than the previously developed video-generating AI Make-A-Video.

Meta launched Emu Edit to edit content. This other generative model allows you to define changes made to the clip created by Emu Video. The user can use this to add or remove elements from a scene, change the clip speed, change the ambience, and much more.

The results are just as impressive as the original video, but now even more true to user requests. “Unlike many existing generative AI models, Emu Edit follows instructions exactly, ensuring that pixels in the input image unrelated to the instructions remain intact,” Meta explains.

For research purposes only

Emu Video and Emu Edit are significant advances in the development of generative artificial intelligence. but not yet publicly available. For now, the project is for research purposes only, but Meta recognizes that “the potential use cases are clearly evident.”

Meta is aware that new tools do not replace artists, animators and professionals in this field, but are technologies that can make their jobs easier.