On May 14th, DeepMind CEO Demis Hassabis took theGoogle I/O stageto unveil what he calls Google’s “most capable” genitive video model, Veo.
Veo, according to Hassabis, creates high quality 1080p videos from text, videos, and prompts. Users can also edit the video in post-production using further text prompts along with “capturing the details” of cinematic styles like an arial of landscapes or a time-lapse, as it’s trained toconvert input text into output video.
7 wild OpenAI Sora results from early access testers
OpenAI’s Sora turns text into videos, and artists are taking it to wild places. Check out their crazy, surreal, and sometimes weird results.
Veo strives for that consistency by building upon previous architecture from Video Poet, Lumiere, and more to improve quality and output resolution.
Hassabis further explained that video, unlike other image multi-language models (MLMs), have to not only understand where an image should be in space, but maintain consistency overtime. According to Hassabis, Veo strives for that consistency by building upon previous architecture from Video Poet, Lumiere, and more to improve quality and output resolution.
Showing off the future of video
To show off Veo’s potential abilities, actor, rapper, and director Donlad Glover made a surprising appearance. “Everybody can become a director, and everyone should be,” Glover reflects, as he believes outlets like Veo are really “about storytelling.”
Users who want to start using Veo can join the experimental model – VideoFX’s – waitlist on Google labs starting today.
I install these 13 essential apps on any new Pixel phone
The Google Play Store is filled with high-quality Android apps – from Pulsar to Yuka, these are some of my non-negotiables.
I was asked to put in any information about or link to any relevant work, but at this time am not sure if this will impact getting a spot on the waitlist.
