Google has unveiled its latest video-generation AI tool, named Veo, at its Google I/O live event. Veo is described as offering “improved consistency, quality, and output resolution” compared to previous models.
Generating video content with AI is nothing new; tools like Synthesia, Colossyan, and Lumiere have been around for a little while now, riding the wave of generative AI's current popularity. Veo is only the latest offering, but it promises to deliver a more advanced video-generation experience than ever before.
To showcase Veo, Google recruited a gang of software engineers and film creatives, led by actor, musician, writer, and director Donald Glover (of Community and Atlanta fame) to produce a short film together. The film wasn't actually shown at I/O, but Google promises that it's “coming soon”.
As someone who is simultaneously dubious of generative AI in the arts and also a big fan of Glover's work (Awaken, My Love! is in my personal top five albums of all time), I'm cautiously excited to see it.
Eye spy
Glover praises Veo's capabilities on the basis of speed: this isn't a deletion of human ideas, but rather a tool that can be utilized by creatives to “make mistakes faster”, as Glover puts it.
The flexibility of Veo's prompt reading is a key point here. It's capable of understanding prompts in text, image, or video format, paying attention to important details like cinematic style, camera positioning (for example, a birds-eye-view shot or fast-tracking shot), time elapsed on camera, and lighting types. It also has an improved capability to accurately and consistently render objects and how they interact with their surroundings.
Google DeepMind CEO Demis Hassabis demonstrated this with a clip of a car speeding through a dystopian cyberpunk city.
It can also be used for things like storyboarding and editing, potentially augmenting the work of existing filmmakers. While working with Glover, Google DeepMind research scientist Kory Mathewson explains how Veo allows creatives to “visualize things on a timescale that's ten or a hundred times faster than before”, accelerating the creative process by using generative AI for planning purposes.
Veo will be debuting as part of a new experimental tool called VideoFX, which will be available soon for beta testers in Google Labs.