The headlining occasion of Google I/O 2025, the live keynote, is formally within the rearview. Nonetheless, when you’ve adopted I/O earlier than, you could know there’s much more taking place behind the scenes than what you could find live-streamed on YouTube. There are demos, hands-on experiences, Q&A periods, and extra taking place at Shoreline Amphitheatre close to Google’s Mountain View headquarters.
We have recapped the Google I/O 2025 keynote, and given you hands-on scoops about Android XR glasses, Android Auto, and Project Moohan. For these within the nitty-gritty demos and experiences taking place at I/O, listed here are 5 of my favourite issues I noticed on the annual developer convention right this moment.
Controlling robots along with your voice utilizing Gemini
Google briefly talked about throughout its fundamental keynote that its long-term aim for Gemini is to make it a “common AI assistant,” and robotics needs to be part of that. The corporate says that its Gemini Robotics division “teaches robots to understand, comply with directions and regulate on the fly.” I received to check out Gemini Robotics myself, utilizing voice instructions to direct two robotic arms and transfer object hands-free.
The demo is utilizing a Gemini mannequin, a digital camera, and two robotic arms to maneuver issues round. The multimodal capabilities — like a stay digital camera feed and microphone enter — make it straightforward to regulate Gemini robots with easy directions. In a single occasion, I requested the robotic to maneuver the yellow brick, and the arm did precisely that.
It felt responsive, though there have been some limitations. In a single occasion, I attempted to inform Gemini to maneuver the yellow piece the place it was earlier than, and shortly discovered that this model of the AI mannequin does not have a reminiscence. However contemplating Gemini Robotics remains to be an experiment, that is not precisely shocking.
I want Google would’ve centered a bit extra on these purposes through the keynote. Gemini Robotics is precisely the type of AI we should always need. There is no want for AI to exchange human creativity, like artwork or music, however there’s an abundance of potential for Gemini Robotics to remove the mundane work in our lives.
Attempting on garments utilizing Store with AI Mode
As somebody who refuses to attempt on garments in dressing rooms — and hates returning garments from on-line shops that do not match as anticipated simply as a lot — I used to be skeptical however excited by Google’s announcement of Store with AI Mode. It makes use of a customized picture era mannequin that understands “how completely different supplies fold and stretch in line with completely different our bodies.”
In different phrases, it ought to provide you with an correct illustration of how garments will look on you, fairly than simply superimposing an outfit with augmented actuality (AR). I am a glasses-wearer that often tries on glasses just about utilizing AR, hopeful that it will give me an thought of how they will look on my face, solely to be disenchanted by the outcome.
I am pleased to report that Store with AI Mode’s digital try-on expertise is nothing like that. It shortly takes a full-length photograph of your self and makes use of generative AI so as to add an outfit in a approach that appears shockingly reasonable. Within the gallery under, you possibly can see every a part of the method — the completed outcome, the advertising photograph for the outfit, and the unique image of me used for the edit.
Picture 1 of 3
Is it going to be excellent? Most likely not. With that in thoughts, this digital try-on software is definitely one of the best I’ve ever used. I might really feel far more assured shopping for one thing on-line after attempting this software — particularly if it is an outfit I would not usually put on.
Creating an Android Bot of myself utilizing Google AI
Loads of demos at Google I/O are actually enjoyable, easy actions with so much of technical stuff occurring within the background. There is no higher instance of that than Androidify, a software that turns a photograph of your self into an Android Bot. To get the outcome you see under, a fancy Android app circulate used AI and picture processing. It is a glimpse of how an app developer would possibly use Google AI in their very own apps to supply new options and instruments.
Androidify begins with a picture of an individual, ideally a full-length photograph. Then, it analyses the picture and generates a textual content description of it utilizing the Firebase AI Logic SDK. From there, that description is distributed to a customized Imagen mannequin optimized particularly for creating Android Bots. Lastly, the picture is generated.
That is a bunch of AI processing to get from a real-life photograph to a customized Android Bot. It is a neat preview of how builders can use instruments like Imagen to supply new options, and the excellent news is that Androidify is open-source. You possibly can be taught extra about all that goes into it here.
Making music with Lyria 2
Music is not my favourite medium to include AI, however alas, the Lyria 2 demo station at Google I/O was fairly neat. For these unfamiliar, Lyria Realtime “leverages generative AI to provide a steady stream of music managed by consumer actions.” The thought is that builders can incorporate Lyria into their apps utilizing an API so as to add soundtracks to their apps.
On the demo station, I attempted a lifelike illustration of the Lyria API in motion. There have been three music management knobs, solely they have been as large as chairs. You could possibly sit down and spin the dial to regulate the proportion of affect every style had on the sound created. As you alter the genres and their prominence, the audio taking part in modified in actual time.
The cool half about Lyria Realtime is that, because the identify suggests, there is no delay. Customers can change the music era instantly, giving folks that are not musicians extra management over sound than ever earlier than.
Producing customized movies with Move and Veo
Lastly, I used Flow — an AI filmmaking software — to create customized video clips utilizing Veo video-generation fashions. In comparison with primary video turbines, Move is used to allow creators to have constant and seamless themes and kinds throughout clips. After making a clip, you possibly can change the video’s traits as “components,” and use that as prompting materials to maintain producing.
I gave Veo 2 (I could not attempt Veo 3 as a result of it takes longer to generate) a difficult immediate: “generate a video of a Mets participant hitting a house run in comedian model.” In some methods, it missed the mark — certainly one of my movies had a participant with two heads and none of them truly confirmed a house run being hit. However setting Veo’s struggles apart, it was clear that Move is a useful gizmo.
The flexibility to edit, splice, and add to AI-generated movies is nothing wanting a breakthrough for Google. The very nature of AI era is that each creation is exclusive, and that is a foul factor when you’re a storyteller utilizing a number of clips to create a cohesive work. With Move, Google appears to have solved that drawback.
Should you discovered AI speak throughout the primary keynote boring, I do not blame you. The phrase Gemini was spoken 95 occasions and AI was uttered barely fewer on 92 events. The cool factor about AI is not what it could actually do, however how it could actually change the best way you full duties and work together along with your units. To this point, the demo experiences at Google I/O 2025 did a strong job at displaying the how to attendees on the occasion.