💥 Google Just NUKED the AI Scene with Gemini Ultra, Veo 3, Imagen 4 & More! 🚀🤖 | AI Revolution - video Dailymotion

Ai Revolution

Google drops a bombshell in AI innovation with the launch of Gemini Ultra, Veo 3, Imagen 4, and other powerful new models. These breakthroughs redefine AI performance, creativity, and multimodal capabilities, setting a new gold standard for the industry. From ultra-fast processing to stunning image generation, Google is reshaping the future of artificial intelligence like never before. 🌟🔥  #GoogleAI #GeminiUltra #Veo3 #Imagen4 #ArtificialIntelligence #AINews #TechBreakthrough #MachineLearning #DeepLearning #AIRevolution #NextGenAI #MultimodalAI #Innovation #FutureTech #AIUpdate #GoogleInnovation #AIModels #TechNews #Superintelligence #DigitalRevolution #AIAdvances

Transcript

00:00Google just went full beast mode at I.O. 2025.

00:06Massive AI upgrades, a $250 Ultra plan that thinks before it speaks,

00:12full-on filmmaking tools with sound and video stitched together by AI,

00:17a search tab that books your tickets for you,

00:19and glasses that turn real life into a live Gemini demo.

00:24We've got robots that code, models that generate apps in seconds,

00:283D video calls that feel like teleportation,

00:31and a new VO3 model that makes AI movies with background noise, music, and actual dialogue.

00:37This isn't an update. It's a full reset of Google's entire ecosystem.

00:41So let's talk about it.

00:43First off, Google set the stage with numbers that are almost cartoonish.

00:46A year ago, they processed 9.7 trillion tokens a month.

00:51Right now, they're chewing through over 480 trillion, 50 times more.

00:567 million developers are already building with Gemini,

00:59and the consumer app has blown past 400 million monthly active users.

01:04Dunder Pichai's phrase was shipping at a relentless pace, and the graphs prove it.

01:09The average ELO score across their models is up 300 points since the original Gemini Pro,

01:14and 2.5 Pro now sweeps every category on the LM Arena leader.

01:20All that is running on the new Ironwood TPU pods,

01:2410 times the performance of the last generation,

01:27maxing out at 42.5 exaflops per pod.

01:31So yeah, they're basically bragging that the hardware is no longer the bottleneck.

01:35On the consumer side, the headline is the Gemini Ultra subscription, $249.99 a month,

01:43United States only for now.

01:45Although, if you're a first-time subscriber, Google gives you 50% off for the first three months.

01:51So it starts at around $125 a month before jumping to full price.

01:55That Ultra badge unlocks VO3 video generation with native sound effects and dialogue,

02:01the Flow filmmaking workspace,

02:03the new DeepThink reasoning mode inside Gemini 2.5 Pro,

02:07bigger limits in Notebook LM,

02:10the Whisk Image Remix tool,

02:12plus YouTube Premium,

02:13and 30 terabytes of Google storage.

02:16If $20 felt steep for the old Gemini Advanced tier,

02:20$249 sounds insane,

02:23until you realize Ultra is their all-you-can-eat compute buffet.

02:27A single VO render with spatial audio can burn more GPU minutes

02:31than most indie developers use in a week.

02:33So Google is basically asking,

02:35are you in or out?

02:37DeepThink itself is worth pausing on.

02:40Regular Gemini 2.5 Pro was already strong,

02:43but it answered in one pass like GPT-3RA models.

02:46Flip on DeepThink and it runs a parallel chain of thought,

02:49evaluating multiple solution paths before it speaks.

02:52That extra reflection time crushes the math and coding benchmarks

02:56that OpenAI's O1 Pro and O3 Pro had been flaunting.

03:00Right now, DeepThink is limited to trusted testers through the Gemini API,

03:05and Google is running extended safety checks before they open the floodgates,

03:09but we'll be benchmarking it the second that toggle appears in studio.

03:13Everyone wanted to see new media models and Google delivered two.

03:17VO3 is the headline grabber,

03:19capable of generating 30-second full high-definition clips

03:23with improved physics and for the first time synchronized audio generated on the fly.

03:29That means footsteps, ambient noise, and even bits of dialogue come built in.

03:37They left behind a ball today. It bounced higher than I can jump.

03:42What manner of magic is that?

03:45It's a major leap towards cinematic quality AI video.

03:55Then there's Imagen 4, focused on still images,

03:59and it's all about precision, capturing textures like fabric,

04:03water droplets, and animal fur with impressive clarity.

04:06Google also mentioned that a new variant is on the way

04:09that could be up to 10 times faster than Imagen 3.

04:12Both of these models plug directly into Flow,

04:15Google's new filmmaking interface,

04:17where users can chain scenes together,

04:19extend clips, and blend reference images.

04:21It's not fully polished yet,

04:23especially when it comes to mixing elements from different models.

04:27But it finally gives multimodal creation

04:30a workspace that feels more like editing than guesswork.

04:34Now, speaking of AI upgrades you can actually build with,

04:37here is something big.

04:38DeepAgent just made something huge possible.

04:41You can now create your own version of ChatGPT

04:43and embed it directly into your website or app.

04:46This update turns DeepAgent into a full platform

04:49for building custom AI chatbots that feel personal,

04:52useful, and totally under your control.

04:55You get to choose the model,

04:56whether it's GPT, Gemini, or another top-tier LLM.

05:00And you can customize everything from the theme and personality

05:04to the exact data your chatbot pulls from.

05:07Want it connected to your Google Drive,

05:09SharePoint, website docs, or even live internet sources?

05:12No problem.

05:13With the new model context protocol integration,

05:15DeepAgent makes it simple to hook your bot

05:18into the tools and content you already use.

05:21This means you can create an AI chatbot

05:24that acts like a therapist, a customer support rep,

05:27a financial advisor, or even a fun digital persona.

05:31And unlike basic plugins, this one lives on your site,

05:35under your branding, with your design.

05:38It's like having a mini ChatGPT running on your own domain.

05:42DeepAgent can also build dashboards, generate documents,

05:45automate workflows, and even interact with platforms

05:48like Google Tasks, Slack, Jira, and GitHub.

05:51All this is packed into a clean interface

05:53that lets you deploy your bot and app instantly

05:57and manage everything in one place.

06:00If you've ever wanted to build your own smart assistant

06:03or AI agent that actually knows your business or project,

06:06this is it.

06:07DeepAgent just turned every website

06:10into a potential AI-powered experience.

06:13All right, now, back to Google I.O.

06:17The live assistant story got louder, too.

06:19Gemini Live now rolls out camera and screen sharing

06:22for every iOS and Android user this week,

06:25powered by the low-latency Project Astra stack.

06:28You can chat naturally, flip the camera around,

06:30and the model keeps up in near real-time.

06:32Google showed it grabbing directions from maps,

06:35dropping events into calendar, and filling to-dos in tasks

06:38without ever leaving the call.

06:40If that ties into personal context,

06:43if you grant permission,

06:45Gemini can mine your Gmail threads,

06:47drive docs, even past itineraries,

06:49then draft a reply that sounds like you.

06:52In the demo, it answered a friend's road trip question

06:55matching the sender's casual greeting,

06:58pulling exact campsite links from an old spreadsheet,

07:01and even mirroring favorite word choices,

07:03all while promising the whole flow is private

07:06and under your control.

07:08We'll see how that plays once the privacy watchdogs weigh in.

07:11Search got a double upgrade.

07:14AI overviews already serve 1.5 billion users,

07:17but Google just flipped on a dedicated AI mode tab

07:20for everyone in the United States starting today.

07:23Regular queries still show classic links,

07:26yet one hop over you get a conversational answer

07:28with sources, follow-ups, and, in a few months,

07:31live data visualizations for sports and finance.

07:34During the demo typing a dense NBA,

07:37StatsQuestion produced its own chart on the spot,

07:40no third-party plug-in required.

07:42Project Mariner's web action chops are sliding into that tab too.

07:46Ask for baseball tickets and AI mode can navigate the team site,

07:50pick seats, and hand you a checkout button already filled out,

07:54all while you watch from the side panel.

07:56Google swears the agent stays under your control,

07:59but the dream is obvious.

08:01Skip the blue links,

08:02let Gemini buy the thing.

08:04Speaking of Mariner,

08:05developers gained an SDK hook to those computer use capabilities,

08:10and early testers like UiPath

08:12are teaching it repetitive back-office tasks.

08:15The neat trick is teach and repeat.

08:18Show the agent one full workflow,

08:20and it generalizes the plan for similar jobs later.

08:23Regular users on Ultra will see that same muscle inside the Gemini app as agent mode.

08:28Think apartment hunting.

08:29Give it the wishlist.

08:30Three bedrooms in Austin,

08:32washer-dryer,

08:33$1,200 each,

08:35and it pings Zillow,

08:37adjusts filters,

08:38schedules a tour,

08:39and reports back,

08:40all while you chill.

08:42On the collaboration front,

08:44Google Meet absorbed Beam,

08:46the artist formerly known as Project Starline.

08:49The hardware still rocks,

08:51a six-camera array

08:52and custom light field display for 3D telepresence.

08:55But now there's AI-driven,

08:57near-perfect millimeter head tracking

08:59and 60-frame video.

09:01More jaw-dropping is live speech translation

09:03that keeps the original speaker's voice, tone, and facial expressions.

09:07English-Spanish hits beta first for AI Pro and Ultra subscribers,

09:12and Enterprise Workspace customers can request early testing later this year.

09:18Developers didn't leave empty-handed.

09:20Stitch debuted as an AI front-end designer,

09:22describe the layout,

09:23or even upload a mock-up,

09:25and it spits back HTML and CSS you can tweak.

09:29Android Studio picked up journeys and an agent mode

09:32to walk through complex build steps,

09:34plus crash insight analysis powered by Gemini,

09:38jewels the coding agent graduated to handling GitHub pull requests

09:42and backlog tickets,

09:43setting itself up as a head-to-head rival

09:45to OpenAI's code interpreter-style workflows.

09:49Meanwhile,

09:50Google AI Studio now exposes the lightning-fast Gemini Flash model

09:54and will add the new Imogen endpoint

09:57once the servers stop melting.

10:00A quick sweep of the smaller but still noteworthy

10:03launches.

10:04Wear OS 6 introduces unified fonts on tiles

10:08and dynamic theming that syncs watch face colors with pixel hardware.

10:12Google Play gets topic browse pages for movies and shows,

10:16United States only for now.

10:18Audio samples so you can preview in-app content,

10:21and a new checkout flow with multi-product subscription bundles.

10:24Subscription add-ons finally live under one payment umbrella,

10:27and developers can kill a live release if a fatal bug shows up in the first hour.

10:32Huge quality of life fix for hardware, Gemma 3N,

10:36a 4 billion parameter model optimized for phones, laptops, and tablets arrives in preview

10:42with full multimodal support.

10:44And yes,

10:45Synth Ida Detector is now a public portal.

10:48Upload an image, audio file, text, or video,

10:51and it flags whether Google's invisible watermark is embedded.

10:55That's going to be essential as VO content starts flooding social feeds.

10:59Infrastructure fans got one more geeky nugget, Gemini Diffusion,

11:04an experimental text-to-application model that uses parallel generation

11:08to spit out functional prototypes basically instantaneously.

11:11They demoed it, generating an entire front-end app in the time it took to narrate the prompt.

11:16That same parallel technique underpins the new Flash model,

11:20which is second only to 2.5 Pro in capability,

11:23but wins on speed and cost, landing generally in early June.

11:27There's a hardware cherry on top.

11:29Project Astroglasses morph into Android XR.

11:33During the live demo, the presenter asked Gemini through the lenses

11:37to remind them of the coffee shop name printed on their cup,

11:40then overlay walking directions in full 3D.

11:44Samsung, Warby Parker, and Gentle Monster are official partners,

11:48so by the time Meta's next Ray-Ban collaboration ships,

11:51Android will have its own XR ecosystem waiting.

11:55Now all of this inevitably begs the price question.

11:58Google's tiering is pretty clear.

12:00The everyday crowd gets AI overviews,

12:02Gemini Live Voice and baseline image generation for free.

12:05The $20 AI Pro plan, formerly Gemini Advance,

12:08grabs you 2.5 Pro.

12:10Standard VO and Imogen and larger context windows.

12:13Ultra at $249.99 is where the bleeding edge toys live.

12:17VO3 with audio, 30 terabyte storage.

12:20Deep think, flow, agent mode, massive 30,000 page context bucket.

12:25Plus the experimental developer knobs like Mariner, Teach and Repeat.

12:30Europeans, yes it's a headache right now.

12:32VPNs and billing addresses still trip the upgrade flow,

12:37but Google promises wider rollout soon.

12:40We'll see.

12:41The subtext through all these launches is that Google is cannibalizing its own classic products.

12:46Chrome will get a Gemini sidebar that summarizes any page.

12:49Searches AI mode threatens the blue link economy.

12:53Play Store topic pages gently steer users away from third-party recommendation blogs.

12:58And with Beam and LiveMeet translation, those standalone virtual events platforms lose a major selling point.

13:05Google's betting that owning the full vertical from TPU silicon to consumer UI will fend off competition from OpenAI, Anthropic, and whoever else rolls up with a flashy demo.

13:20As always, the proof will come when real users slam these tools at scale.

13:25Will VO3 stay coherent on a 10-second camera pan?

13:29Does deep think hallucinate less or just hallucinate more confidently?

13:34Can SynthID survive a heavy Instagram filter?

13:37Over the next few weeks, I'll be stress testing Ultra, pushing deep research on 50 academic PDFs,

13:43teaching Mariner how to file expense reports, and seeing if those personalized Gmail replies actually sound like me or like corporate copy.

13:52That's the whirlwind tour.

13:54Trillions-scale token counts, parallel thinking language models, 3D telepresence, AI-built apps in a blink,

14:00and a subscription tier that costs more than some people's rent.

14:04Google didn't just iterate this year.

14:06It carpet-bombed the whole product line with generative AI.

14:10The ball is squarely in OpenAI's court now.

14:14Well, thanks for watching, and I'll catch you in the next one.

14:18It's okay.

14:20It's okay.

14:22It's okay, we'll catch you in the next two weeks.

14:24It's okay.

14:26I'm going to catch you next time.

14:28It's okay.

14:30You're looking to make it for a second.

14:32It's okay.

💥 Google Just NUKED the AI Scene with Gemini Ultra, Veo 3, Imagen 4 & More! 🚀🤖 | AI Revolution

Category

Transcript

Recommended