Google on Monday unveiled a successor to the Veo video generation model. Dubbed Veo 2, the artificial intelligence (AI) model improves on the capabilities of the older model which was slated to be launched in early 2024. Veo 2 can now generate video at up to 4K resolution and for much longer durations. The new model was introduced just a week after the Mountain View-based tech giant added VO to Vertex AI for its enterprise customers. Veo 2 is currently only available to select creators on the VideoFX platform through Google Labs.
Google launches Veo 2 with improved capabilities
In a blog post, the tech giant explained in detail about the new video-generation AI model. Currently, the older Veo AI model is available to users through Vertex AI and VideoFX, and is also being used in the video background generator feature for YouTube Shorts. Now, the company has introduced Veo 2, which is said to have achieved cutting-edge results during internal testing, beating video models like Sora.
Introducing VO2, our new, state-of-the-art video model (with a better understanding of real-world physics and motion, up to 4K resolution). You can join the waiting list at VideoFX. Our new and improved Imagen 3 model also achieves SOTA results, and is arriving in over 100 countries today… pic.twitter.com/FtwNKyicEc
– Sundar Pichai (@sundarpichai) 16 December 2024
The tech giant claimed that Veo 2 has improved the details and realism of the generated videos. Additionally, the AI model can now understand the language of cinematography. This means users can specify a style, lenses, cinematic effects and camera movements and Veo 2 can produce videos to exact specifications.
Highlighting an example, the post states that the VO2 can understand complex signals such as “a low-angle tracking shot with an 18mm lens that passes through the middle of a scene”. The mention of the lens will let the AI know that the user wants to generate a wide-angle shot, which the lens is known for, the company said.
Additionally, Veo 2 can generate video up to 4K resolution and extend video duration up to “minutes”. The company has not specified the maximum duration of videos that can be created using the AI tool. The tech giant also claimed that the video model reduces the incidence of hallucinations and that unwanted details such as extra fingers, deformed limbs, or unexpected objects will be seen less frequently.
On security, Google said the company has added the native SynthID watermark on generated videos to help identify AI videos from real videos, reducing its misuse to create deepfakes. The company also plans a gradual rollout of the AI model to control its access and improve its quality and security before global availability.
Currently, Veo 2 is available to select users in VideoFX through Google Labs. Those interested in trying the tool are asked to sign up for the waiting list. Additionally, the tech giant plans to expand VO2 to YouTube Shorts and other products next year.
Imagen 3 also launching globally at ImageFX
Along with the launch of Veo 2, the tech giant also launched Imagen 3, an AI image generation model in ImageFX, to more than 100 countries. The latest iteration of the model has received several upgrades, including the ability to render more diverse art styles, greater accuracy, and better-composed images.
The company claimed that the Imagen 3 exhibits better instant adherence and can render more detail and texture in images. It is also said to excel in specialized styles such as photorealism, impressionism, abstract, and anime.
