The Google I/O 2024 keynote was a 112-minute long affair where the company made several big announcements focused on artificial intelligence (AI). The announcements ranged from new AI models to the integration of AI into Google products, but perhaps one of the most interesting unveilings was Veo, an AI-powered video generation model that can generate 1080p video. The tech giant said the AI tool can generate videos that exceed the one-minute mark. Namely, OpenAI also presented its video AI model called Sora in February.
During the event, Demis Hassabis, co-founder and CEO of Google DeepMind, introduced Veo. Announcing the AI model, he said: “Today I’m excited to announce our newest and most capable generative video model Veo. Veo creates high quality 1080p videos from text, images and video queries. It can capture the details of your instructions in a variety of visual and cinematic styles.”
The tech giant claims Veo can follow instructions closely to understand the nuance and tone of a phrase and then generate a video to match it. The AI model can generate videos in different styles such as timelapse, close-ups, fast tracking shots, aerial shots, and different shots with lighting and depth of field. In addition to generating videos, the AI model can also edit videos when a user provides it with an initial video and a prompt to add or remove something. Furthermore, it can also generate videos longer than one minute, either through a single response or through multiple consecutive queries.
To solve the consistency problem in video generation models, Veo uses latent diffusion transformers. This helps reduce instances where characters, objects, or the entire scene unexpectedly flicker, jump, or change between frames. Google emphasized that videos created by Veo will be watermarked using SynthID, the company’s internal AI-generated content watermarking and identification tool. The model will soon be available to select creators through the VideoFX tool in Google Labs.
Ve’s similarities to OpenAI’s Sora
Although neither AI model is yet available to the public, both share several similarities. Veo can generate 1080p videos lasting over a minute, while OpenAI’s Sora can generate videos up to 60 seconds. Both models can generate videos from text queries, images and videos. Based on the diffusion model, both are capable of generating videos from multiple shots, styles and cinematographic techniques. Both Sora and Veo also come with AI-generated content tags. Sora uses the Coalition for Content Provenance and Authenticity (C2PA) standard, while Veo uses its native SynthID.
