Google introduces text-to-video AI model Veo to take on OpenAI's Sora

A screenshot of a video created by Google Veo (Image credit: Google)

Google on May 14 introduced Veo, its most advanced video generation model that can generate high-definition video in a range of cinematic and visual styles, at its annual developer conference Google I/O 2024.

This announcement comes amidst intensifying competition in the artificial intelligence (AI) video generation models space from rivals such as OpenAI's Sora, Facebook parent Meta's Emu Video, and startups such as Runway, and Stability AI.

Sora, in particular, has wowed people with its realistic visuals since its debut in February.

Veo generates high-quality 1080p resolution videos that can go beyond a minute. The company claims the model has an advanced understanding of natural language and visual semantics, and can generate video that closely represents the user's creative vision — accurately rendering details in longer prompts and capturing tone.

It also understands cinematic terms like "timelapse" or "aerial shots of a landscape" and can create consistent and coherent footage where people, animals and objects move realistically throughout shots. One can further edit these generated videos using additional prompts.

"We are also exploring features like storyboarding and generating longer scenes," said Demis Hassabis, the CEO of Google DeepMind.

Building on Google’s earlier AI video generation efforts

Veo builds upon Google’s years of efforts in generative video models, including Generative Query Network (GQN), DVD-GAN, Imagen-Video, Phenaki, WALT, VideoPoet and Lumiere - combining their architecture, scaling laws and other techniques to improve quality and output resolution.

The company said it is currently inviting a range of filmmakers and creators to experiment with the model. These collaborations will help the company improve how it designs, builds and deploys these technologies, with a goal to ensure that creators "have a voice in how they are developed", it said.

During the conference, Google also previewed its collaboration with filmmaker Donald Glover and his creative studio, Gilga, who experimented with Veo for a film project.

"With Veo, we’ve improved techniques for how the model learns to understand what's in a video, renders high-definition images, simulates the physics of our world and more." said Eli Collins, Vice President, Product Management at Google and Doug Eck, Senior Research Director, Google in a blogpost.

"These learnings will fuel advances across our AI research and enable us to build even more useful products that help people interact and communicate in new ways" they said.

Veo will be available to select creators as a private preview inside Google's AI video generator, VideoFX, which is part of the company's Labs initiative, over the coming weeks. People can sign up to join the waitlist. Google said it will also bring some of Veo's capabilities to YouTube Shorts and other products in the future.

New Imagen 3 model

Google also introduced a new version of its image generation model Imagen, which the company says is its highest quality text-to-image model to date.

Called Imagen 3, the model can produce photorealistic, lifelike images with incredible level of detail and far fewer distracting visual imperfections than prior models, the company executives said in the blogpost.

Imagen 3 better understands natural language, the intent behind the user's prompt, and also incorporates small details from longer prompts, they said.

"It’s also our best model yet for rendering text, which has been a challenge for image generation models. This capability opens up possibilities for generating personalized birthday messages, title slides in presentations and more" said Google executives Collins and Eck.

Imagen 3 will be available to select creators as a private preview inside Google's AI image generator ImageFX through its Labs initiative. People can sign up to try the model by joining the waitlist. Imagen 3 will be soon available to developers and enterprises through Vertex AI, the managed AI app developer platform of Google's Cloud unit.

Apart from these developments, Google announced that Grammy winning musician Wyclef Jean, electronic musician Marc Rebillet and Grammy nominated songwriter Justin Tranter have released demo song recordings created with the help of the company's music AI tools on their YouTube channels.

The tech giant is also expanding its watermarking tool SynthID, that embeds digital watermarks AI-generated images and audio, to more formats including text and video. All videos generated by Veo on VideoFX will also be watermarked by SynthID, the company said.

Event alert: Moneycontrol and CNBC TV18 are hosting the ultimate event on artificial intelligence, bringing together entrepreneurs, ecosystem enablers, policymakers, industry leaders, and innovators on May 17 in Gurugram. Click here to register and gain access to the AI Alliance Delhi-NCR Chapter.