Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Google and/O 2025. It never concerned the subtlety. This year, the company has given up incementalism, providing a cascade of generative upgrades AI that aims to draw a search ticket, video and digital creativity.
Linchpin: Gemini, Google’s Family Family, now powers everything, from search results to video synthesis and creating a high-resolution image-exhibiting new territory in the race everything defined with what is quickly and how originally, AI can generate.
It’s Showstopper I see 3Google’s first AI video generator that creates not only visual displays, but also complete sound records – clear buk, effects, even dialogue – they combined directly with the shots. Includes the text and the pictures are entering, and the exit fully produced 4K video.
This indicates the first large video model that can generate audio and visual content-trend that has started with at the same time Showrunner alphaUnpublished model, but VEA3 offers far versatility, generating different styles outside the simple 2D animation of cartoon films.
“We are entering a new era of creation with a combined generation of audio and videos,” said VP Google Labs VP Josh Woodward while starting. It is a direct challenge for the current leaders of the video generation generation-Kling, Hunyuan, Luma, Wan and Openai’s Sora-which placed the veil as a solution in one, not demanding more tools.
In addition to the Veli3, Imagen 4 – the latest itteration of the image generators – leads with improved photorealism, 2K resolution, and perhaps most importantly, by textual display that actually acts for inscriptions, products and digital wet.
For all who have suffered a motor text created by previous AI models of image, Imagen 4 represents a significant improvement.
These tools do not exist isolated. Flow AI, a new feature of professional users subscriptions, combines linguistic veil capabilities, imaded and geminia into a combined environment to create a movie and edit the scene. But this integration comes at the price – $ 125 a month for access to the complete tool as a part of the promotional period until the price of a full $ 250 starts to charge.
Gemini: Starting searching and “text diffusion”
Generative AI is not just for creators of content. The twins 2.5 now forms the backbone of the redesigned company’s search engine, which Google wants to evolve from the aggregate of the link to the dynamic, conversational interface that processes complex inquiries and delivers synthesized answers to several sources.
AI review – where Google Gemini tries to provide comprehensive answers to inquiries without users require them to click on other websites – now they sit on top of the search page, and Google has reported more than 1.5 billion monthly users.
Another interesting development is “Gemini diffusion”, built with the technology he started The founding of the laboratory a few months ago. Until recently, the Ai community generally agreed that authorgressive technology was best functioning to create text, while diffusion technology was excellent for images.
Autoregressive models generate every new token after reading all previous generations to determine the best token – the ideal for making coherent answers of the text constantly inspecting the prompt and the previous exit.
Diffusion technology works differently, starting with fulfilling all contexts by accidental information and refining (diffuse) output each step to make the final product match the prompt – perfect for pictures with fixed payments and aesthetics.
Openai first successfully applied an authorial generation to the image models, and now Google has become the first large company to apply a diffusion generation to the text. This means that the model begins with nonsense and perfects the whole effect for every iteration, creating thousands of tokens per second, maintaining accuracy – for context, groq (not Xai’s Grok), which is one of the fastest requirements in the world, generates almost 275 tokens in speed, and the traditional providers cannot be approached.
The model, however, is not yet publicly available – Interesting users must Join the waiting list– But early adoptive parents shared impressive results that show the speed and precision of the model.
Practically with Google’s AI tools
We got our hands on several Google new AI features, with mixed results depending on the layer.
Deep research is particularly strong – even beat the Chatgpt alternative. This comprehensive investigative agent estimates hundreds of sources and provides reliable information with minimal mistakes.
What gives him an advantage over Openi’s research agent is the ability to generate infographics. After creating a complete research text, this data can condense in visually attractive diamonds. We fed the model all about Google’s latest announcement, and he presented accurate information through the charts, schemes, charts and cards of mind.
Veli 3 remains exclusive to Gemini Ultra users, although some third-party providers like Freepik and Fal.ai already offer access to API. The flow is not available to try out if you do not sprout an ultra plan.
Flow has proven to be an intuitive video editor with Vewing models at their core, allowing users to decorate, cut, expand and change AI scenes using simple text inquiries.
However, even the VEH2 had little love, making life easier for professional users. Generations with the now affordable VEA2 have been significantly faster-we have made 8 seconds of video in about 30 seconds. Although the VEK is missing sound and currently supports only the text-Video (with the image-video-video image soon, he understood our instructions and even generated coherent text.
VE2 is already compared to Kling 2.0 – observing the reference value of quality in the generative video industry. New generations with Veli3 look even more realistic, more coherent, with good background sound and life dialogue and voices.
It is difficult for the Imagena to determine at first glance whether Google is installing version 4 or still uses version 3 on their Gemini Chatbot interface, although users can confirm it via Whisk. Our initial tests suggest that Imagen 4 is prioritized by realism, unless otherwise stated, with better fast adherence and visuals that outweigh their predecessor.
We generated a picture with different elements that usually do not fit into the same scene. Our inquiry was “A photo of a woman with a skin made of glass, surrounded by thousands of glossy and essential pieces in a baroque room with the word” decipherpipe “written in neon, realistic.”
Although both Imagen 3 and Imagen 4 understood the concept and elements, Imagen 3 failed to capture a realistic style – which Imagen 4 easily did. Generally, Imagen 4 is comparable to the generators of the Sota image, especially given how easy it is to accelerate.
Sound examinations have also improved, and the models have now easily provided more than 20 minutes full of twins’ discussions, instead of forcing users to transfer to a notebook. Because of this, the twins are a more complete interface, reducing fragmentation that has previously demanded that users jump between different web sites for different services.
The quality is comparable to the one of Notebookwith slightly longer outings on average. However, the key feature is not that the model is better, but that it is now embedded in Gimini’s Chatbot User interface.
Premium AI at premium price
Google did not hide its entry strategy. Company “Ultra“The plan costs $ 250 a month, a priority priority approach to the most powerful models, a flow of Alans and 30 storage terabyets – pure aiming of filmmakers, serious factories and businesses. $ 20 AI Pro” is free to unlock the Google model, with pictures, similar and paintings, on basic tools. Token Cap and only 10 studies per month.
This layered approach to mirrors wider trend of market AI: Encourage mass adoption with free, then lock professionals with features that are too useful to be conveyed. Google is a bet that a real action (and margin) is in superior creative work and automated business flows not only occasional instructions and generation meme.
Edited Andrew Hayward
Generally intelligent Bulletin
Weekly AI journey narrated by gene, generative AI model.
