Google’s new AI video model is less bad at physics

Google may have only recently started rolling it out I understand generative AI to corporate clientsbut the company wasted no time in releasing a new version of the video tool for early testers. On Monday, Google announced a preview of Veo 2. According to the company, Veo 2 “understands the language of cinematography.” In practice, this means that you can specify a specific film genre, cinematic effect or lens when suggesting the model.

Additionally, Google says the new model has a better understanding of real-world physics and human movement. Correctly modeling people in motion is something that all generative models struggle with. So the company’s claim that the Veo 2 is better when it comes to both of these pain points is remarkable. Of course, the samples provided by the company aren’t enough to know for sure; the true test of the Veo 2’s capabilities will come when someone prompts it generate a video of a gymnast routine. Oh, and speaking of things video models struggle with, Google says the Veo will produce artifacts like extra fingers “less often.”

Separately, Google is rolling out improvements to Image 3. Of its text-to-image model, the company says the latest version generates brighter and better-composed images. Additionally, it can render a wider variety of art styles with greater accuracy. At the same time, it is better to follow the prompts more faithfully. Fast adherence was an issue I highlighted when the company made Imagen 3 available to Google Cloud customers earlier this month, so if anything, Google is aware of the areas where its AI models need to work.

Veo 2 will gradually roll out to Google Labs US users. For now, Google will limit testers to generating up to eight seconds of footage at 720p. For context, Sora it can generate up to 20 seconds of 1080p footage, though it costs $200 a month ChatGPT Pro subscription. As for the latest Imagen 3 enhancements, they are available to Google Labs users in more than 100 countries via ImageFX.