Google responds to OpenAI's launch of Sora.
Google's DeepMind unveiled the Veo 2 model on Monday, a video generation system that can produce clips of up to two minutes in length and in 4K resolution.
The Google DeepMind team has recently unveiled its video generation model Veo 2, which represents a significant advancement compared to its previous version. This new model is capable of creating clips of up to two minutes in length and with a 4K quality resolution, which is six times longer and four times higher resolution than the 20-second clips at 1080p that its predecessor, Sora, could generate. However, these figures are theoretical limits, as currently, Veo 2 is only available on VideoFX, an experimental video generation platform by Google, where clips are limited to eight seconds and a resolution of 720p.
Additionally, access to VideoFX is restricted; therefore, not all users can immediately try Veo 2, although the company expects to expand access in the coming weeks. A Google spokesperson stated that Veo 2 will also be usable on the Vertex AI platform once its capabilities can be properly scaled. Eli Collins commented that the company will continue to refine the model based on user feedback and aims to integrate the new capabilities of Veo 2 into various use cases within the Google ecosystem, anticipating sharing more updates next year.
The Veo 2 model stands out for offering a better understanding of physics, achieving more realistic lighting effects and fluid dynamics. It also produces sharper video clips, with more defined textures and images that are less blurry during movement. Among its additional features are enhanced camera controls, allowing users to position the virtual lens with greater precision. However, there are still areas that require improvement, such as consistency in adhering to complex instructions over extended periods, as well as the creation of intricate details and complex movements.
On the other hand, Google also announced improvements to its image generation model, Imagen 3, which can now create brighter and better-composed results. This update will also include descriptive suggestions based on keywords from user prompts, displaying menus of related terms for each given keyword.
