Advertisement
From the jellyfish pre-show to the “TPU Training Day” film, discover how Gemini played a crucial role in making I/O 2026 a reality.
Your browser does not support the audio element.
Google I/O 2026 focused on how AI is being made helpful for everyone in innovative ways. The event didn’t just showcase AI advancements — the event itself was brought to life using these AI tools.
Advertisement
This year marks a unique and thrilling time to create, with AI tools rapidly improving and changing the creative landscape. The challenge was to use these same AI tools, showcased on stage, to push the limits of innovation, creativity, and efficiency.
We accelerated our processes, prototyping in real-time, and blending human artistry with cutting-edge technology, best exemplified by the “Timmy TPU” film.
The real reward lies in showing how these tools can unlock creativity and take over mundane tasks, allowing the team to focus more on their uniquely human contributions. When executed well, the event stands out on its own, making viewers forget about the AI involvement. This transformation is what we aim to convey, addressing the common question, “What can you really do with AI?”
Read on to find out which AI tools we employed and how we used them to bring I/O 2026 to life.
The AI products & models: Google AI Studio; experimental DeepMind models; Gemini Omni; Nano Banana
What we did: We produced a short film featuring TPUs preparing for the heavy lifting at I/O 2026.
How we did it: The project began with a simple question: Could we animate a film using basic materials like cardboard and markers, then use AI to animate it? Collaborating with director Laurie Rowan and Nexus Studios, we combined puppetry, traditional animation, and AI to keep human creativity at the forefront of “TPU Training Day,” also known as “Timmy TPU.”
We began by capturing puppet performances and simple 3D animations, providing full control over framing and camera angles. We used Nano Banana to generate stylized initial frames from this raw footage. A custom tool was developed within Google AI Studio to ensure consistency and pixel-perfect matches before creating sequences.
We combined the base animation with the stylized frames using Gemini Omni and other experimental models, elevating the film to a cinematic level while preserving the human touch. These small imperfections give puppet films their charm, and our AI pipelines were designed to maintain these details.
The AI products & models: Gemini models and Nano Banana
What we did: We developed the visual brand identity for I/O 2026, featuring a four-color gradient with overlapping transparencies and interlocking icons.
How we did it: Our brand identity was crafted through collaboration between our team and AI. We began by inputting our past brand guidelines and five years of I/O recaps into Gemini models. Early outputs didn’t fully meet our expectations, so we conducted micro-experiments. We generated new imagery, iteratively refining it with Nano Banana feedback. We also explored icon styles, ultimately choosing flat 2D icons that transform into hyper-textured 3D icons, creating a cohesive brand across keynotes, signage, and apps.
The AI products & models: Google Antigravity; Google Colab; Google CoralNPU; Google Flow Music; Lyria 3 Pro
What we did: We launched the pre-show with Jellectronica, a generative music experiment with the Monterey Bay Aquarium, translating moon jelly movements into sound using Lyria 3 Pro.
How we did it: We trained a YOLO8 model in Google Colab, running it on Google’s Coral NPU to track jellyfish movements and control music generated with Google Flow Music and Lyria API. For instance, more jellies in the bass section increased the bass’s volume and energy. We also coded a mass stem generator in Google Antigravity to automate the creation of music stems like bass, chords, melody, and drums.
The AI products & models: Google AI Studio; Gemini API; Gemini Canvas; Google Antigravity; Lyria 3; Nano Banana
What we did: Infinite Scaler, another part of the pre-show, was a video game where players generated levels as they played.
How we did it: We aimed for players to quickly build infinite 3D worlds using 2D image generation. We used Nano Banana to create sprite sheets from user prompts and references via the Gemini API, and processed foreground elements to generate depth maps. These textures were then mapped to a 3D cardboard box in WebGL. Google AI Studio was used for rapid prototyping, transitioning to Google Antigravity for full development, and Lyria 3 generated the in-game music.
Players can explore the levels we built together by playing the game.
The AI products & models: Flutter; Gemini Enterprise Agent Platform; Google Antigravity; Nano Banana
What we did: An app was created for I/O attendees to design and order lattes with custom art, enabling them to build their own coffee app version.
How we did it: We employed generative UI and the A2UI protocol with Flutter to create adaptive interfaces that updated in real-time, replacing static forms with dynamic interactions. Firebase connected the frontend to models like Nano Banana, managing complex reasoning and content generation. A single Flutter codebase ensured seamless user experiences across various devices. Google Cloud and Firebase, including Cloud Functions, Firestore, and Cloud Ops, handled the complexities of modern generative AI apps. Attendees also used Google Antigravity’s coding capabilities to quickly develop their own order apps.
The AI products & models: Gemini Omni; Google Flow; Nano Banana Pro
What we did: Custom generated title cards were made for each speaker with our image and video generation models.
How we did it: Josh Woodward, VP of Google Labs & Google Gemini, was featured digitally riding Chrome Dino and dunking a basketball. Nano Banana Pro generated core assets such as ingredient reference sheets, which were used to storyboard various actions and details. Google Flow’s Veo assisted in prototyping and animating actions like a slam dunk. Consistent AI outputs were ensured with detailed prompts, and raw motion from generated videos was polished into final titles.
The AI products & models: Gemini and Nano Banana
What we did: Custom I/O stickers were instantly generated and printed for attendees.
How we did it: An interactive sticker game was developed on a custom web app, where players caught falling prompts in 20 seconds using an Android bot. With over 100 prompt categories, players could choose two prompts or hit “I’m feeling lucky” for a random combination. The backend, utilizing Nano Banana for Gemini and Android, fused these choices into personalized sticker designs, instantly printed for attendees to collect.
Learn more about these and other major I/O 2026 announcements here.
