Google has introduced a new and improved Veo 2 video generator model to compete with the likes of OpenAI’s Sora. The company claims that the successor to the original Veo AI model can create realistic motion and high-quality output of up to 4K, which it says is better than leading AI video generator platforms. Alongside this, Google also announced the latest Imagen 3 version and new Whisk model to create a single image from multiple visuals. Here is everything to know. Google shared a series of short video clips created using Veo 2, which shows that this platform can generate hyper-realistic videos of animals and food. We can also see animated clips of humans, all of which are 8-second videos. “Veo 2 outperforms other leading video generation models, based on human evaluations of its performance,” Google said. While the company hasn’t mentioned the names of the rivals, it is likely pointing towards OpenAI’s Sora — which is also a video generator. In the benchmark list, the company has added a graph, which claims that its Veo 2 model is preferred more by people in comparison to Meta Movie Gen, Kling V1.5, Minimax and Sora Turbo.
The samples shared by Google look great, but some scenes with motions are seemingly a bit inaccurate. A few details in parts of a frame are missing. Google acknowledges this and says that complete consistency throughout complex scenes or those with complex motion still remains a challenge. But, the overall quality of the videos is seemingly quite impressive. “While Veo 2 demonstrates incredible progress, creating realistic, dynamic, or intricate videos, and maintaining complete consistency throughout complex scenes or those with complex motion, remains a challenge. We’ll continue to develop and refine performance in these areas,” said Google DeepMind.
As for the Imagen 3 model, Google claims it can now create brighter and more realistic images with vibrant hues, better colour balance, and fidelity. The company is also claiming that it can currently produce highly detailed textures and attractive visuals.