Google is rolling out a new way to generate videos, as well as animate existing images, using AI. The catch? These videos are really short, and you have to pay to play.
On Tuesday, Google announced that Gemini Advanced and Google One AI Premium users can now make AI-generated videos using the company's Veo 2 video model. Google originally unveiled this model back in December, touting the model's improved understanding of real-world physics, nuanced human emotions, and ability to generate specific types of shots (e.g. low angle, close-up, shallow depth of field).
While Veo 2 is capable of generating 4K videos at several minutes in length, that's not the experience Gemini Advanced users will get here. Instead, you'll be able to make eight-second videos at 720p—quite the limitation. Speaking of limits, there's a cap on the number of videos you can generate with Veo 2 each month, though Google isn't disclosing that publicly. Instead, the company says it'll warn you when you're approaching your limit, which doesn't seem like a particularly transparent system.
While Gemini Advanced users will be able to use Veo 2 directly in Gemini, Google One AI Premium subscribers also have access to the model in Whisk—the company's AI media generator. As part of the Veo 2 rollout, Google is announcing Whisk Animate, which uses Veo 2 to animate an image you generated with Whisk.
How to generate Veo 2 videos with Gemini Advanced
If you pay the $20 per month for Gemini, Veo 2 is rolling out right now. To start, open Google's chatbot, then pick "Veo 2" from the model dropdown. From here, prompt the AI like you would normally, only this time with a specific video in mind. While Google encourages users to describe things like short stories for Veo 2 to work with, remember that you only get eight seconds of video generation max. That's quite the short story.
If you don't see Veo 2 as an option, sit tight. Google says the full rollout could take a few weeks, so not all Gemini Advanced users will be able to access it today.
How to animate a Whisk image with Veo 2
Google One AI Premium subscribers can give this one a shot over at labs.google/whisk. You can start by generating an image with Whisk as per usual—by prompting Whisk with whatever you want, or letting the AI randomly pick a generation for you.
Once your image generates, you'll notice a new "ANIMATE" option at the top. Click this, then prompt the AI with how you'd like it to animate the image.
What's the point of this?
I don't have a subscription, so I can't try out Veo 2 in any capacity. But according to Google's demos, the model does a decent job of adhering to the prompts. For example, Google showed off how four different versions of a man eating ice cream—including 3D art, pixel art, illustration, and realistic generation—could be animated with a single prompt.
That said, it's far from perfect: Each of these demonstrates the hallmark signs of AI-generated video, including visual glitches and elements morphing into one another. Some of the results here are also half-baked: The 3D art animation, for example, has the man bring the ice cream to his mouth, but never actually eat it, while the realistic generation shows him eating from his fingers, even though there is no ice cream there.
There's a novelty to the feature, sure: You can generate a brief animation of any image you want—AI-generated or not. But I struggle to imagine the practical and widespread application here, especially since this feature is locked behind Google's $20 per month paywall.
