Google’s new AI video model sucks less at physics

Google may have only recently started to launch its I see Generative AI at the company’s customersBut the company does not waste time to get a new version of the video tool to the first 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 refer to a specific genre of film, cinematic effect or lenses when submitting the model.

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

Separately, Google has implemented improvements Figure 3. Of its text-to-image model, the company says the latest version generates brighter, better-composed images. Also, it can render more different art styles with more precision. At the same time, it is even better to follow the suggestions more faithfully. Ready adherence was an issue I highlighted when the company made Imagen 3 available to Google Cloud customers earlier this month, so if nothing else, Google is aware of the areas where its AI models need to work.

Veo 2 will roll out gradually Google Labs users in the United States. For now, Google will limit testers to generating up to eight seconds of 720p footage. For context, Sora it can generate up to 20 seconds of 1080p footage, although it costs $200 per month. ChatGPT Pro Subscription. As for the latest Imagen 3 improvements, those are available to Google Labs users in more than 100 countries through ImageFX.