Start building and exploring worlds of your own with Project Genie. Project Genie is an experimental research prototype that lets you create and explore infinitely diverse worlds. Simply use a text or image prompt to generate interactive environments that build in real time as you play them. Project Genie is powered by Genie 3, our world model which enables training AI agents in an unlimited curriculum of rich simulations and opens new frontiers for AI research. It’s currently available to Google AI Ultra s ...

[music] Hi, I'm Davi, a product manager at Google DeepMind covering genomics. I'm joined [music] by JGA, who leads our genomics team, Natasha, Tom, and Jun. We've just released our recent work, Alpha Genome, in Nature.Alpha Genome is a unified DNA sequence [music] to function prediction model that predicts the functional impact of genetic variance. And today we wanted to talk about the science behind it. Uh why we [music] built it, how it came about, um what we released, and what we're excited about next.Ma ...

Heat. Heat. ...

I've been working animated films like Pixar's Inside Out 2 for the last 8 years. My name is Connie Hu and I'm the director for The Ox Neighbors. For this film, the creative challenge was to bridge the gap between art and technology.For the visuals, we really wanted to bring in abstract textures and styles. My father is an artist and his painting inspired aesthetics we captured on screen. But hand painting 7,000 framed in such a complex style wasn't feasible.We had to scale the process without losing that hu ...

Gemini 3 Flash excels in capabilities required for rapid prototyping, including instant code generation, aesthetic versatility, and creative ideation. It can also natively drive the design process, like generating new prompts for style exploration, and can produce multiple distinct UI variations with just a single click to accelerate development. Learn more at https://deepmind.google/models/gemini/flash/ ___ Subscribe to our channel https://www.youtube.com/@googledeepmind Find us on X https://twitter.com/Go ...

What we have here is a simple web app that takes simulated unstructured data, in this case a chat, and matches it to the right customer record, bringing everything together. It's a good example of how Gemini 3 Flash can quickly pull information from different systems and merge it into one clear, organized piece of data. ...

Gemini 3 flash enables an entirely new way to develop. It's low latency and quality coding capabilities allow you to AB test real-time code generations. In this example, Gemini 3 flash evolves the perfect loading spinner.So fast, you barely need a loading spinner. Once you start evolving, it streams the code for the next version of spinners and shows you the latency metrics. You select your favorite variant and the next generation of spinners will be based upon that one.You can imagine this for any app, per ...

Gemini 3 Flash excels in speed and efficiency by coding SVG images, HTML, and three.js side-by-side between 3 Flash and 2.5 Pro. 3 Flash generates the imagery better, faster and with fewer token usage. Learn more at https://deepmind.google/models/gemini/flash/ ___ Subscribe to our channel https://www.youtube.com/@googledeepmind Find us on X https://twitter.com/GoogleDeepMind Follow us on Instagram https://instagram.com/googledeepmind Add us on Linkedin https://www.linkedin.com/company/deepmind/ ...

Gemini 3 Flash specializes in capabilities critical for live assistance, including complex geometric calculation, velocity estimation, and multimodal processing. It can also simultaneously analyze video and hand-tracking inputs to provide strategic, real-time guidance in this slingshot game. Learn more at https://deepmind.google/models/gemini/flash/ ___ Subscribe to our channel https://www.youtube.com/@googledeepmind Find us on X https://twitter.com/GoogleDeepMind Follow us on Instagram https://instagram.co ...

Gemini 3 Flash can handle a huge number of function calls reliably at low latency. In this function calling challenge, it must successfully sequence tasks to prepare dishes. Let's cook some ramen.We start with 100 ingredients and 100 tools which the model has to reason across simultaneously. Each kitchen tool is a separate function call. There are three agents powered by Gemini 3 flash.The cooking agent triggers the function calls. It plans ahead by combining raw ingredients to use later on. New combination ...