Google recently revealed a demo trailer for their new Lumiere AI, an AI-powered tool designed to generate videos from simple text prompts. The software was developed by the team at Google Research.
Inbar MosseriInbar, Team Lead and Senior Staff Software Engineer at Google Research announced on X (formerly Twitter), “Thrilled to announce “Lumiere” – the new text-to-video model we’ve been working on! Lumiere generates coherent, high-quality videos using simple text prompts.”.
See Related: WIN NFT HERO from TRON’s Metaverse Gears Up for the GameFi Stage
Capabilities Of Lumiere
As well as a research paper, the company also released a trailer video showcasing some of the capabilities of the new model. The AI is capable of generating “realistic, diverse and coherent motion” from texts such as “a dog driving a car wearing funny glasses”. Additionally, Lumiere can also make videos from existing photos, using texts as guidelines.
Google also demonstrates the AI’s ability for stylized generation, where it uses any photo as a reference and creates a video in the same art style.
In the research paper, Google claims its model is superior to existing video generation models as it uses “Space-Time U-Net architecture that generates the entire temporal duration of the video at once”.
At the time of writing, Google’s Lumiere is not available to the public. Interested parties can find samples of its work on Lumiere’s GitHub page.