Google’s latest creation, Veo, is a cutting-edge AI model capable of producing 1080p video clips up to a minute long based on a text prompt.
Unveiled at the Google I/O 2024 developer conference, Veo can capture various visual and cinematic styles, such as landscapes and time lapses, and can also make edits and adjustments to existing footage.
Demis Hassabis, the head of Google’s AI R&D lab DeepMind, mentioned during a virtual roundtable that they are exploring features like storyboarding and generating longer scenes to test the capabilities of Veo. He expressed pride in the progress they have made in the realm of video generation.
Veo represents an advancement from Google’s previous work in video generation, as demonstrated in April, which utilized Imagen 2 models to create looped video clips.
In contrast to the Imagen 2-based tool, which could only produce low-resolution, short videos, Veo is now competitive with leading video generation models like Sora and those from startups such as Pika, Runway, and Irreverent Labs.
Douglas Eck, the lead researcher at DeepMind in generative media, showcased some impressive examples of Veo's capabilities, including a vibrant beach scene that showcased its superiority over other video models.
Veo’s training involved extensive footage, a common method for generative AI models to learn patterns in data and generate new content, in Veo’s case, videos.
When asked about the source of the training footage, Eck hinted that some may have been sourced from YouTube. He emphasized that any data sourced from YouTube was governed by agreements with creators.
Google’s updated terms of service have enabled the company to leverage more data for training its AI models, potentially including data from YouTube. However, Eck maintained that Google is committed to ethical practices in utilizing training data.
While Google joins other tech giants in utilizing user data for training AI models, the company asserts that it is prioritizing ethical considerations in its approach.
As Veo is made available to select creators, including Donald Glover and his creative agency Gilga, Google is positioning Veo as a tool for creative professionals.
Creators concerned about data scraping from YouTube and issues like regurgitation in generative AI remain wary, as Google does not provide mechanisms for creators to remove their work from training data sets post-scraping.
Eck highlighted Veo's technical capabilities, including its ability to understand camera movements, VFX, and physics, contributing to the realism of the videos it generates.
Despite its impressive features, Veo is not without flaws, such as objects disappearing and reappearing inconsistently. It will be available on Google Labs for the foreseeable future, with plans to integrate its capabilities into products like YouTube Shorts.
As Veo continues to evolve, Google aims to enhance its video creation and editing capabilities for a broader audience of creators, positioning it as a tool for innovative filmmaking.
We're launching an AI newsletter! Sign up here to start receiving it in your inboxes on June 5.
