Did you know you can customize Google to filter out garbage? Take these steps for better search results, including adding my work at Lifehacker as a preferred source.
OpenAI’s new Sora app has been the main focus concerning hyper-realistic AI slop over the past few weeks. Sora makes it all too easy for users to generate short-form videos that seem real enough to most people, including videos that showcase real people’s likenesses.
But before Sora dropped, it was Google that was stirring fears about these realistic AI videos. With Veo 3, Google launched an AI model that not only produced lifelike videos, it also generated realistic audio synced to the action. Sound effects, environments, even dialog could all generate alongside the video itself, selling the effect entirely from one simple prompt.
Veo 3.1
Now, Google is back with an upgrade to Veo, appropriately named Veo 3.1, which the company announced in a blog post on Wednesday. This isn’t necessarily an overhaul or a revolutionary new video model. Instead, Veo 3.1 builds on top of Veo 3, adding “richer audio” and “enhanced realism” that Google says generates “true-to-life” textures. The new model also reportedly supports new narrative controls tools, which pairs with new upgrades to Flow, Google’s AI video editor. Flow users now have more granular controls when editing, and can add audio to existing features like “Ingredients to Video,” “Frames to Video,” and “Extend.”
What does that mean in practice? According to Google, Ingredients to Video with Veo 3.1 lets users add references images to their scenes, such as a specific person, clothing items, or an environment. The new Flow editor can then insert those elements into the finished product, as you can see in the demo video below:
Building off of this new feature, Flow now lets you add new elements to an existing scene as well. With “Insert,” you can tell Veo 3.1 to add new characters, details, lighting effects, and more to the clip. Google says it is also working on the opposite as well, to allow users to remove any elements they don’t like from a generation.
Google also now has a new way for users to dictate how they’d like a scene to generate, called “First and last frame.” Users can choose reference frames for the beginning and ending of a scene. Flow with Veo 3.1 will then fill in the gap, and generate a scene that starts and ends based on those images.
What do you think so far?
There’s also now a way to create videos that are longer than previous iterations of Flow would generate. The new “Extend” feature lets you either continue the action of the current clip, or cut to a new scene that follows it, though Google says the feature is most useful for generating a longer establishing shot. According to the company, Extend can create videos that last over a minute.
Veo 3.1 is available for users in the Gemini app as well as Vertex AI, as long as you have a Google AI Pro subscription. Developers can access it via the Gemini API. Google says Ingredients to Video, First and last frame, and Extend are coming to Gemini API, but “Add object” and “Remove object” are not available. “Extend” is also not yet available in the Vertex AI API.
Is this really a good thing?
Google sees all of these advancements as a boon for creatives and creativity, but I’m highly skeptical. I could see Veo 3.1 and Flow as a good tool for envisioning shots before filming or animating them (i.e. a storyboarding tool), or even a way for new and budding filmmakers to learn editing by seeing their ideas in a more realized form. However, as a whole, I don’t think AI-generated content is the future—or, at least, not a future most of us want. Sure, there’s humor or novelty in some of these AI-generated videos, but I’d wager most of the people who enjoy them do so ironically, or exclusively to social media.
The idea of replacing human filmmakers and actors with AI generations seems absurd, especially when it puts all of us at risk of disinformation. Is it really so important for companies like Google and OpenAI to make it easy to generate hyper-realistic fully-rendered scenes, when those videos could so easily be used to trick the masses? This could be the ramblings of someone resistant to change, but I don’t think most of us would like to see our favorite shows and movies made with passion and emotion, replaced by realistic-looking people delivering muted and robotic performances.
