Highlights:
- Google has released a generative AI video model called Veo that can create high-quality, photorealistic 1080p video clips.
- Veo supports text-to-video, video-to-video, and image-to-video conversions for a wide range of cinematic styles.
- Google partnered with artist Donald Glover to test Veo's new features, showcasing amazing video generation capabilities.
Webmaster's Home (ChinaZ.com) May 15 News: Google researchers at its deep learning AI unit, DeepMind, have released a brand new AI video model called Veo, capable of creating "high-quality, 1080p clips of more than 60 seconds," which "can cope with a range of cinematic styles, from photorealism to surrealism and animation," reaching an astonishing level of realism and visual results.
Veo's goal is to help people of all kinds create video, whether it's an experienced filmmaker, an aspiring creator, or an educator eager to share knowledge.
Veo supports text-to-video, video-to-video, and image-to-video conversions for all film styles, from realism to surrealism and animation.
Google partnered with artist Donald Glover to test some of Veo's new features through his startup studio, Gilga. DeepMind has released a number of V-generated videos and tips on YouTube and X platforms, including neon cities, real ocean jellyfish, cowboy horseback riding, spaceships traveling through the void, and real character scenes, among others. These videos are almost indistinguishable from live-action or professional, computer-generated animations, and are all generated by text prompts.
The image comes from Google, and the official video screenshot was generated with Veo
Veo can not only generate videos based on text prompts, but also quickly edit AI-generated video user-uploaded clips or even pre-recorded live footage. When given an input video and editing command, such as adding a kayak to an aerial shot of a sea line, Veo can apply this command to the initial video and create a new edited video. As a result, Veo was also able to achieve consistency between video frames, avoiding some of the strange and disturbing transition artifacts thanks to its advanced latent diffusion transformer technology that reduces these inconsistencies and keeps characters, objects, and styles in their place in real life.
To improve the quality of the generated videos, Google added more detail to each video title in the practice data and used high-quality, compressed representations of the video (also known as latent variables) to increase efficiency. In addition, all Veo videos are embedded with SynthID, Google's Content Credential tracking digital watermark to ensure that they can be recognized by recognizable agencies as AI-generated.
Veo is the culmination of years of research at DeepMind and builds on previous research findings, including Generative Query Network (GQN), DVD-GAN, Imagen-Video, Phenaki, ALT, VideoPoet, and Lumiere, among others. At the moment, Google doesn't release Veo publicly, only for a few specific creators to use in private previews. In the future, Google also plans to bring some of Veo's features to YouTube Shorts and other products.