Google Unveils Lumiere, AI Video Generator That Converts Text to Complex Animated Scenes
-
Google has announced Lumiere, an advanced new AI video generator that converts text to video. It can generate scenes and animations based on text prompts or images.
-
Lumiere uses a new "Space-Time Diffusion Model" to generate the entire video at once for better consistency, unlike other models that build videos frame-by-frame.
-
Features include stylizing videos by turning people into toys, creating cinemagraphs with partial animation, and video inpainting to transform sections of a video.
-
The model was trained on a dataset of 30 million 5-second, 128x128 videos and their captions, but details of the training data are unclear.
-
There is no word yet on if or when Lumiere will be publicly launched as a product, despite looking very impressive in demos.