wild what you can do with Veo 3 and Gemini 2.5 Flash Image Preview in Google AI Studio - testing out here with a public domain song, a picture of me, and music styles decade by decade

👋 Want to make sure that your favorite IDEs and extensions (Roo Code, Cursor, Windsurf, etc.) always use the latest Gemini API models and features? Make sure to either add one of the files below into your context window, or MCP-ify the SDK repo: 🔗 JS/TS: lnkd.in/g8nn7GVT 🔗 Python: lnkd.in/g4XKNF-j

🍌 There's always money in the banana stand! Make sure to check out our latest model in Google AI Studio - Gemini 2.5 Flash Image Preview. Editing, reimagining, extraction and more. 🖼️: "The Green Cap" by Alex Katz

a poem about ai art assembled by an ai which composed news headlines about ai art gist.github.com/dynamicwebpa...

small-batch, artisanal, performance-enhancing beverage handcrafted from the purest COCO dataset, with subtle flavor grad-ients 🥥😂

📸 Imagen 4 Ultra in Google AI Studio is extremely, *extremely* good at generating images that look and feel like they were taken with a Polaroid camera. Test it out yourself: aistudio.google.com

📽️ Simple research and podcast generation workflow using LangGraph and Gemini 2.5! You can pass a research topic and a YouTube video; the system will then perform research on the topic using search, analyze the video, combine insights, and generate a report with citations as well as a short podcast.

Google AI Studio + @github.com = 🚀! You can now link the apps that you create with your GitHub (personal or private repos). Happy building!

Reminder: DeepMind's Gemma 3n model is performing about as well as Gemini 1.5 Pro – better, in many cases! – and is only 4B parameters in size. The best model 6 months ago is now small enough to be run on a laptop. Now play that forward 6 months from now, thinking about the best models of today. 🤯

Get ready to enter the simulation... Genie 3 is a new frontier for world models: its environments remain largely consistent for several minutes, with visual memory extending as far back as 1min. These limitations will only decrease with time. Welcome to the future.🙌 deepmind.google/discover/blo...

📝 Tired of having too many tabs open, but no time to read through all of them? Just built a Chrome extension using AI Studio's URL Context feature - all you have to do is add an API key, and the extension uses Gemini 2.5 Flash Lite to summarize all of your open tabs. 🔗 github.com/dynamicwebpa...

🚀 The Gemini APIs in Google AI Studio can be used to understand – and also to generate! – many different output modalities: Video, images, audio, text, code, or all of the above all at once. If you've only been using the models to generate text and code, you're missing out. 👇 ai.google.dev

🏗️ More problem sets like this one from @mitocw.bsky.social, please: "The intent of this first civil engineering design project is to let you get a feeling for [material properties and constraints] -- by going from playing to building and assessing *what* you do and *why*."

Standing room only for the @majorleaguehacking.bsky.social + Google DeepMind + Google Cloud AI Roadshow in San Francisco! "This is the single greatest moment in history to become a software engineer, or a technologist – to be anyone who is creating digital artifacts. And we're only just beginning."

🗺️ New favorite use case for Google AI Studio's Build: 1️⃣ Use Gemini w/Search Grounding to identify specific places (ex: dog-friendly parks in SF; pickleball courts that are bookable on weekends; aestheticians that use Dermalogica products; etc.) 2️⃣ Then use @openstreetmap.bsky.social to visualize.

"In collaboration with the @archive.org TV News Archive we examined 14,749 evening news broadcasts spanning the big three networks from July 2010 to present, asking Gemini 2.5 Flash Thinking to make an index of the stories covered in each broadcast, geography, sentiment, frame, narrative structure.

🥳 The @majorleaguehacking.bsky.social + Google DeepMind AI Roadshow is happening tomorrow in SF! Six hours of learning and experimenting with the latest Gemini and Veo models in Google AI Studio, seeing what others have built, great food and drinks, and even better company. 🔗 lu.ma/SFroadshowML...

Just finished presenting on the Gemini APIs, Google AI Studio, Gemma 3 and 3n, @googlecolab.bsky.social's Data Science Agent, and @googlelabs.bsky.social's Jules to a group of educators and engineers at Stanford. Thanks for having @developers.google.com and Google Cloud come visit today! 🌲 ☁️

👋 Series A founders: Want to work directly with DeepMind and Google Cloud engineers, have a fast track to support, and understand how others are using AI at their companies? Apply for the Google for Startups Gemini Founders Forum, a two-day, in-person event in the Bay Area. Apply by Sept. 10th!

obsessed with the prufrock bunny 🐇☕

✨ 🙌 Grateful to spend this morning and afternoon with ~100 founders from London, as part of the Google for Startups AI sprints! Projects included everything from healthcare, to finance, to marketing and environmental science - all shipped in minutes using aistudio.google.com + the Gemini APIs.

comparison of the present-state of agentic capabilities in @github.com Copilot, @sourcegraph.com Cody, @cursor.com.web.brid.gy, Cline, and several others - I just wish it was a constantly refreshing website instead of a PDF 😅 www.preprints.org/frontend/man...

Interesting comparison between Gemini CLI and Claude Code, across 50+ software development tasks: "These measurements come from real development teams working on production applications and show that tool choice should depend on your specific use case rather than following blanket recommendations."

🏗️ Grateful for Joyce and @nathanbenaich.bsky.social, who shared the surreal, extremely Kubrickesque EDP headquarters in Lisbon last night! For anyone else whose physics brain is breaking, the cantilevered bit over the stairs is an ultra-long balcony -- you can actually walk down it:

❤️‍🔥Gemini 2.5 to build the plan for migrating a codebase, and Claude Code (or Gemini CLI) to execute. Pretty soon the concept of "painful and time-consuming migration from one version of an SDK to another" will feel like as ancient of an idea as punchcards:

😞 So sad to see paperswithcode is discontinued – but grateful that the @hf.co team is, as always, stepping up to support the community! It was an incredible resource for use cases, common themes in papers, and visualizing how models have improved on evals over time: huggingface.co/papers/trend...

Google is processing 980 trillion+ monthly tokens across our products and APIs (up from 480T in May) 🤯 wen googolplex? 🤩❤️‍🔥

"...in which an anonymous poet laments that women of his era know much about artifice and fashion, but little about practical matters such as stewing chiles with cheese." 🧀 🤠 www.foodrepublic.com/2017/10/03/h...

From Prompt to Deployed App in Less Than 2 Minutes, by @dynamicwebpaige.bsky.social: https://dev.to/googleai/from-prompt-to-deployed-app-in-less-than-2-minutes-dh3 #ai #prompting #deploying #google #guides

💖 Loved this Lex Fridman conversation with Demis! His excitement about how our latest video generation model, Veo 3, is learning intuitive physics (ex: fluid dynamics) just from watching videos is contagious... like reverse-engineering the physical world, or reality itself.

🙌 Calling all early-stage founders: DeepMind and Google Cloud are organizing a startup founders forum in November! It's going to be two full days of technical presentations, feedback sessions, prototyping, demos, and insights into product roadmaps. 👇 Apply below: blog.google/outreach-ini...

🦖🍦 Did you know that you can enforce character and scene consistency using just @googlelabs.bsky.social Flow and the Veo APIs? 👇 All you have to do is specify "frames to video" with a character sheet as the first frame, or specify "ingredients to video" for multiple characters + scenes:

Gemini 2.5 Flash-Lite, our fastest and most cost effective model, is now GA and ready for scaled production use!! Make sure to try it out - especially for agent workloads. Comes with native reasoning capabilities, a 1M token context window, and is priced at ($0.10 in / 1M) and ($0.40 out / 1M).

"mouthfeel", except it's the satisfying sound of campfires crackling, crunching snowshoes on fresh powder, waves crashing, tennis balls hitting a racket in the exact center, the smol whirring noise as electric cars amble along i am a woman in want of whatever this word is

🪠One of my cousins recently moved into her very first apartment, so I built a Google AI Studio micro-app that can quickly triage whether a plumbing situation is a "MAYDAY CALL YOUR GRANDPA" or a "MAYDAY CALL A LICENSED PROFESSIONAL". 🧠 Gemini 2.5 📸 Live API 🔍 Search grounding 🔗 lnkd.in/ge-CuHpg

"This achievement is a significant advance over last year. At IMO 2024, AlphaGeometry and AlphaProof required experts to first translate problems from natural language into domain-specific languages, such as Lean, and vice-versa for the proofs. It also took two to three days of computation. [...]

vibes

🦙 ollama mug doing overtime this weekend as a vase for unexpected bouquets 💐🌻

looks like it's about time for the annual Halt and Catch Fire speedrun, folks

confession: i really, really like occasionally pulling all-nighters for work-related or learning or reading-a-book sorts of reasons, and think that the energy an all-nighter brings is unique and irreplaceable

spoilers for my talk at @devrelcon.bsky.social this afternoon 🤫

👀 Want to make sure that your favorite code generation tools (like @windsurfai.bsky.social, @cursor.com.web.brid.gy, @roocode.bsky.social, Cline, @github.com Copilot, and more) use the latest version of the Gemini models? 👇 All you need to do: copy and paste the text below into your context window:

apropos of nothing: this is an appreciation post for @simonwillison.net, who tries out literally all the tools, and joyfully shares (for free!) what he learns in the hope that it will be useful to other humans grateful to have someone relentlessly building with the garage door open 💗🙌

✨🦖 Blown away by Saga, powered by Google's Gemini APIs, as well as Veo 3 and Imagen 4! Used it to generate the first scene for a science-fiction film involving physicists and dinosaurs. Check out this opening footage and frame: 🔗 writeonsaga.com

🤖 Solid performance from Gemini 2.5 Pro and Flash on the new ManipBench robotics dataset! "ManipBench evaluates low-level robot manipulation reasoning capabilities of VLMs across various dimensions, including how well they understand object-object interactions and deformable object manipulation."

👋 See y'all in about 10 minutes! If you've never created an agent before, but want to learn how to get started (and see some examples of production-grade agents in action, via @googlecolab.bsky.social and @googlelabs.bsky.social products!) then this is the @odsc.bsky.social session for you.

claude shannon and hedy lamarr romcom, lfg (generated with ❤️ and Veo via Ingredients-to-Video in Flow)

👋 If you're around #ICML2025 tomorrow and interested in learning more about Google DeepMind + very non-traditional paths in machine learning research: would love to say hi! Demos, puns, and segues into all of the rad experiments that GDM is driving in the natural and physical sciences = guaranteed:

✨👀 If you're interested in getting free GPU credits for your research (on Google Cloud and via other CSPs), and having your feedback help shape the next State of AI Report, then fill out the survey below: <5 minutes, and *really* love the detail in the usage questions: airstreet.typeform.com/survey

goshhhhhdang 🚢🏔️ #ICML2025