Move over, ChatGPT! Gemini’s not just showing up to the AI race, it’s bringing new tricks, sharper tools, and a fresh sense of personality.
Ahead of Google I/O 2025, Google gave us a preview of what’s next for its fast-evolving AI assistant. From real-time visual interaction and podcast-style summaries to lifelike video generation and drag-and-drop code support, Gemini is starting to feel less like a chatbot and more like an all-purpose creative collaborator.
Gemini Live: The AI that sees, hears, and helps
Gemini Live is Google’s bid to make AI feel more like a real assistant and less like a homework helper. Think hands-free voice commands, camera-based context, and real-time feedback, whether you’re baking banana bread or figuring out where that weird museum painting came from.
“That’s the huge power of Gemini Live, being able to be your personal assistant, whether you’re talking to it, showing it something, or doing something and wanting Gemini to provide feedback,” said Elijah Lawal, Google PR Manager.
It’s also multimodal, meaning it sees and hears. You can point your phone camera at your living room and ask for design suggestions, or show it your pre-oven batter and ask if it needs more flour. (Yes, this was an actual demo.)
Audio Overviews: AI podcasts made just for you
One of the sleeper hits of the demo was the “Audio Overview” feature. Take any topic Gemini has researched for you and it can now read it back as a surprisingly chill, podcast-style breakdown.
Think of it as an AI-generated podcast that sounds like two knowledgeable friends casually explaining something over coffee except it’s built specifically for you.
It even handles 45 languages, so you can now listen to your AI-generated think piece about nuclear fusion while folding laundry or commuting.
“We recently increased the length of our audio overviews to give users the opportunity to go even deeper,” said David Sharon, Group Product Manager at Google DeepMind. “It feels like a real conversation, with emotion in the voice. That’s exactly our goal.”
Deep Research: Send Gemini down the rabbit hole
If you’ve ever googled a complicated topic and ended up in 13 tabs of conflicting information, Deep Research is here to spare your brain. It’s like sending your own personal intern to browse the web for 15 minutes, then return with a cited, structured, and digestible report.
It even shows you its research steps and lets you tweak the plan before it starts. Bonus: you can turn the report into an audio episode on the spot.
“It builds a report like a good analyst would, really comprehensive and sourced,” said Aarush Selvan, Senior Product Manager at Google DeepMind. “You can even say, ‘I don’t like step 8’ and it will refine the plan.”
Creative Tools: Images, videos, and dino bedtime stories
Gemini’s creative side is also getting more powerful and more precise.
▪️Image Generation now supports editing, so if your AI dog image needs a birthday hat, you can just tell it.
▪️All generated content is marked with SyntID, an invisible watermark that stays even after external editing.
▪️Veo 2, the AI video generator, can now create surprisingly realistic scenes — like dogs swimming in pools with correct splash physics (yes, really).
“If the water doesn’t ripple right or the dog doesn’t splash naturally, the magic breaks,” Sharon explained. “We worked hard to make it boring because if it’s boring, it means it looks real.”
And yes, it can even write a children’s story about a dinosaur named Ronnie exploring Central Park, complete with illustrations.
Canvas: The “Google Docs for AI”
With Canvas, you can work with Gemini on writing or coding projects in a shared space. Think real-time edits, tone adjustments, auto-suggestions — and even help building interactive web apps, no coding experience required.
In one demo, the team used Canvas to generate a working 3D periodic table… then turned it into a Pokémon-style card game. Nerds, rejoice.
“I studied law and theology. I can’t code,” Lawal admitted. “But now I kind of don’t need to. Gemini just does it for me.”
Personalization, if you want it
By connecting to your Google Search history (only if you opt in), Gemini can give more tailored suggestions. It remembered one demo user liked Kung Fu and anime so it recommended cosplay characters and career switches that matched.
“To be a true assistant, Gemini needs to know a little about you,” said Lawal. “Search history is one way to personalize that experience but you’re always in control.”
It’s weirdly charming. And kind of scary smart.
It’s limited to individual accounts for now. Enterprise accounts don’t have this feature yet.
Trust, transparency, and watermarks
Gemini-generated images and videos now come with SyntID, an invisible watermark embedded directly into the file, one that remains even after editing or third-party exports. This makes it easier to verify AI-generated content in the age of deepfakes, without disrupting creative workflows.
“It’s unbreakable as far as we know,” said Sharon. “Even if someone edits it on a third-party platform and re-exports it, the watermark remains.”
Google also encourages users to double-check AI responses using its integrated “fact check” feature, which cross-references Gemini’s answers with Google Search.
“No LLM is 100% perfect, just like no human is 100% perfect,” Lawal noted. “That’s why we offer the double-check feature, so you can verify what Gemini tells you.”
What’s coming at I/O?
With Google I/O 2025 happening next week, expect Gemini to take center stage. We may see new model announcements, tighter integrations across Google services, or perhaps a look at how Gemini will be built into Android, Workspace, or even future hardware.
“We’re focused on building things people actually want and doing it responsibly.” said Selvan. “It’s not just about keeping up. It’s about pushing the whole space forward.”
If the pre-briefing was any indication, Gemini’s not just here to compete. It’s here to make AI useful, playful, and just a little bit personal.
