Google VEO 3 AI Explained: DeepMinds 8 New AI Video Generator Abilities

Just check out these examples. The microfilm is in your ticket. They’re watching the north exit. Use the service tunnel. Next, there’s native audio generation, where unlike VO2, which generated only silent videos, VO3 finally incorporates all kinds of audio according to text prompt. This includes sound effects, ambient noise, and dialogue that aligns to its visuals. And you can include a serious amount of detail for the audio, just like this. They left behind a ball today. It bounced higher than I can jump. What manner of magic is that? But there’s a caveat, because Google says its audio generation, particularly in instances of short spoken segments, is still being refined, as the synchronization and speech coherence still isn’t perfect.

But VO3 ups the ante, not just with audio, but also with video thanks to feature 7, which is its enhanced prompt adherence. VO3 improves on VO2 by interpreting complex text or image instructions with next-level accuracy, allowing it to follow extremely detailed prompts which unlocks the ability to generate videos with sequential actions and consistent details. Specifically, VO3’s adherence to input prompts uses Google’s natural language processing and scene understanding to create visuals that most closely match the user’s intent. And this is demonstrated in the following examples. But for even more control, there’s a suite of features for video customization, allowing users to control everything from camera angles to movements, as well as framing to achieve practically any kind of effect.

Plus, reference-powered video lets you use input images to define how your scene looks and how it feels, as well as which characters or objects you want, all while keeping consistency throughout. And this allows users to maintain very specific aesthetics for multiple outputs while providing character consistency for unparalleled continuity Plus, VO3 lets users edit video outputs, where outpainting can extend video frames to fit various aspect ratios while maintaining visual coherence. Then, there’s also object manipulation, which allows users to simply add or remove elements, like adding a man with a torch, all while preserving realistic shadows and environmental interactions.

And in other examples, objects can be removed with a prompt. It even features motion master so that users can control the complex movements and paths of objects. Plus, there’s first-to-last frame control for smooth transitions between clips or ideas, and this saves users countless hours of post-production in programs like Blender. But voice motion integration may be the most powerful feature of all, controlling character animation through reference video for consistent facial expressions and voice-driven motion. In fact, users can even record vocal inputs to animate characters’ lip movements and expressions just like this. And while effective in many scenarios, Google still acknowledges that short speech coherence is an area under active development.

Now enter Flow, which is Google’s AI-powered video editing platform. And this supercharges VO3 by allowing users to combine multiple clips, adjust transitions, and even refine audio-visual synchronization for even more polished narratives. And it’s currently available to Google AI Ultra subscribers, as well as enterprise users for Vertex AI Studio. And this integration with VO3 streamlines the workflow from generation to final edit, providing a comprehensive solution for creators. And when it comes to safety, Google has already implemented protocols to VO3 to address misuse risks, where all videos are embedded with Synth ID watermarking to identify AI-generated content.

Additionally, the model blocks prompts involving harmful or copyrighted content and undergoes safety evaluations to mitigate privacy, bias, and infringement issues. Plus, Google is collaborating with its internal teams and external experts to test features before release to ensure responsible deployment. And for Androids, Foundation Robotics Labs just unveiled another sneak peek of its phantom MK1 humanoid robot, which stands at 1.75 meters with a body weight of 80 kilograms. And it’s designed for general purpose tasks. And here it’s seen doing real world pick and place tasks using its cycloid actuators, which deliver up to 160 newton meters of torque with back driving torque of under 1 newton meter.

And the robot’s upper torso features a series of high resolution cameras and sensors, enabling it to process complex data for real time situational awareness for navigating dynamic environments. And it features a payload of up to 20 kilograms for logistics, disaster relief, and more. But when it comes to the robot’s price, not much is known. Although when it comes to robot dexterity, a company named Sharpa just launched Sharpa Wave. And it’s a 22 degree of freedom hand that rivals human precision and speed. And with a fingertip strength of 20 newtons, it combines grasping with rapid dynamic motion.

And it’s designed for lightweight integration with any humanoid robot for collaborative arms, as well as advanced prosthetics and its targeting applications in health care, manufacturing and human to robot interaction. And finally, figure just unveiled a sneak peek of its zero three, marking a significant leap in design. And the company says that it’s going to be optimized for high volume production at the Bot Q facility, where they’re going to produce over 12,000 units annually. And unlike the figure two robot, which was not designed for mass production, figure three is now using tool processes like injection molding and die casting.

And this produces part manufacturing times drastically, in order to reflect a cheaper price when production comes, which they’re targeting at a price point of between 20 to $30,000, most likely to compete with Tesla’s Optimus. But if you want to know more about the latest in AI news, make sure to click this video here, like and subscribe. And thanks for watching. [tr:trw].