A Smarter Memory for Video Continuity

Generating a single AI video clip is one thing. Stringing together several shots that feel like they belong in the same film is a far trickier puzzle. Google thinks it has a better answer with Gemini Omni 1.1 Flash, its latest generative video model now rolling out to developers and subscribers.

The headline upgrade is memory. Instead of only glancing at the final second of footage before continuing a scene, Omni 1.1 can now reference up to 10 seconds of prior video. That extra context makes a real difference. Characters stay looking like themselves. Environments don’t morph into something unrecognizable. The overall direction of a shot holds together, rather than the model guessing wildly about what happens next.

Longer Clips, Real Beginnings and Endings

Your AI videos don’t have to end so quickly anymore. Google now lets creators extend scenes in 10-second chunks, building up to a total length of 40 seconds. That may not sound epic compared to a classic film reel, but it’s a meaningful jump for AI-generated storytelling. Suddenly, you have room for a setup, a conflict, and a resolution.

There’s also a new way to control the shape of a clip. You can supply both a first and a final frame, and Gemini fills in everything between them. That’s handy for constructing a smooth camera move around a subject, zooming between two compositions, or building a seamless loop with no visible jump cut. Video references are supported too. Feed the model up to three seconds of existing footage, and it uses that visual context to keep details like a character’s outfit or hairstyle consistent across generated scenes.

Cheaper Previews Before You Commit

Not every experiment needs to start at maximum quality. One of the more practical additions acknowledges that reality. Developers can generate 360p previews that Google says run up to 60% faster than standard 720p output, at roughly one-third of the cost.

That’s a solid workflow win. You can test an idea, tweak your prompt, and iterate through several versions without burning through your budget. Once you’re satisfied with the direction, Omni 1.1 can render at 1080p or upscale the final result to 4K.

Where to Try It: API, Google Flow, and the Gemini App

Google is making Omni 1.1 Flash available to developers through the Gemini API in Google AI Studio. Businesses can access it via Google’s enterprise Agent Platform.

You don’t need to be building an app to play with these features, though. The model is rolling out globally in Google Flow for AI Plus, Pro, and Ultra subscribers. Scene extension is also coming directly inside the Gemini app for those same subscribers, which makes one of the model’s most interesting new abilities much easier to test on the fly.

For creators already experimenting with AI video generation tools, this update narrows the gap between a rough draft and something close to a finished scene. And if you’re just getting started, the 360p preview option lowers the barrier to entry considerably. It’s a practical step forward in making AI video feel less like a lottery and more like a craft.

Nobody sounds like a news anchor

We all stumble, backtrack, and throw in a dozen “ums” when we talk. Google’s new Gemini 3.5 Transcribe is built for exactly that reality. Announced as the company’s most precise speech-to-text model yet, it takes your messy, unpolished speech and turns it into clean, formatted text — without demanding you sound like a broadcast professional first.

The model arrives alongside two companions, Gemini 3.5 Live and Gemini 3.5 Live Experimental. Together, the trio forms what Google brands as “Gemini Audio.”

Here’s the kicker: it’s already in production, not sitting in some research lab.

How Gemini 3.5 Transcribe cleans up your speech

The model automatically strips out filler words like “um” and “ah,” so your transcribed text doesn’t carry every verbal hiccup. More impressively, it catches self-corrections in real time. Say “let’s meet Tuesday, no wait, Wednesday,” and it understands you meant Wednesday — and adjusts accordingly.

You can also edit your text using nothing but your voice. No typing, no tapping. Just speak the correction.

The model can even delegate tasks like image generation or file analysis to other Gemini models through function calling. That capability is currently live in the Gemini app on macOS.

Accuracy numbers that back it up

Google reports a word error rate of just 4% for real-time streaming and 2.6% for pre-recorded audio. That’s strong for conversational speech, especially when you factor in the messy reality of how people actually talk.

Support spans more than 85 languages, regional accents, and specialized jargon. It can also attribute speech to up to three different speakers in a recording, complete with timestamps. That’s a serious upgrade for journalists, researchers, and anyone who transcribes meetings.

Where it’s already live

This isn’t a future feature. Gemini 3.5 Transcribe already powers Rambler, the dictation feature on Android’s Gboard, and the Gemini app on macOS. Chrome support is coming soon, which will let you dictate directly into any web field.

Developers can access it now in Google AI Studio and Antigravity, Google’s agentic coding platform. Enterprise users get it through the Gemini Enterprise Agent Platform.

Expanding across Google’s ecosystem

The rollout doesn’t stop there. Google is expanding the model into Search Live, Gemini Live, Docs, Keep, and Gmail. The next time you’re rambling into your phone, Gemini might just understand you better than you understand yourself.

For anyone who’s ever dictated a text and gotten gibberish back, this feels like a genuine step forward. The question isn’t whether AI can transcribe — it’s whether it can handle the way we actually speak. Google’s betting the answer is yes.

The Quiet Corner of Apple’s AI Push

Apple’s big AI moment has finally arrived — on the iPhone, iPad, and Mac. But walk into your living room, and the story changes. The Apple TV and HomePod have been sitting this one out, watching from the sidelines. That’s odd, especially since Siri is the main way you talk to both devices.

That could be about to shift. A fresh clue buried in the latest iOS 27 beta suggests Apple is prepping its home devices for a slice of the Apple Intelligence action.

What the iOS 27 Beta Reveals

As spotted by Apple Home Authority, the Home app now includes a new menu labeled “Apple Intelligence Report.” It appears for compatible Apple TV and HomePod units running beta software. A single menu buried in a beta might normally be easy to shrug off. But here’s the thing: neither device currently supports Apple Intelligence at all. So why build a menu for something that doesn’t work yet?

The timing matters too. This isn’t just an iPhone feature accidentally referencing other gadgets. For the menu to show up, your Apple TV or HomePod also has to be running the tvOS 27 or HomePod 27 beta. That’s a deliberate link — software on one device talking to software on another. It points to something real in the works.

Siri Could Finally Catch Up in Your Living Room

The most likely payoff is a smarter Siri. Right now, the experience is uneven. You can ask your iPhone a complex question and get a rich, context-aware answer. Ask the HomePod sitting three feet away the same thing, and you’ll likely get a shrug. That gap has felt awkward for a while, and Apple seems to know it.

Bringing the newer Siri to the HomePod would make voice requests genuinely useful again — especially when your phone isn’t in hand. Imagine asking for a recipe, a weather update, or a reminder without pulling out your device. The Apple TV could benefit just as much. A smarter assistant could help you find something to watch, answer questions about a show, or navigate menus without phrasing every command like a robot command.

There’s a catch, though. None of this is confirmed. Apple hasn’t officially announced Apple Intelligence support for either product. The menu’s existence hints at intent, but it doesn’t tell us which models will actually get the features.

New Hardware Might Be the Missing Piece

Here’s the bigger question: will current Apple TV and HomePod models handle it? Apple Intelligence has steep hardware requirements elsewhere in Apple’s lineup. The iPhone, iPad, and Mac all need relatively recent chips to run it. It wouldn’t be a shock if Apple’s home devices need new silicon too before they can join the AI party.

Rumors have already pointed toward refreshed Apple TV 4K and HomePod models. If new hardware is coming, this beta discovery fits neatly into that timeline. A new chip inside a new HomePod, paired with a smarter Siri, would give the smart speaker a reason to exist beyond just playing music and setting timers.

What About Existing Devices?

Some current models may still get partial support. The menu’s appearance doesn’t reveal which devices are compatible. Apple could limit full Apple Intelligence to new hardware while offering a lighter version to older ones. Or it could leave older devices out entirely, which would frustrate plenty of owners.

We won’t have to guess for long. Apple is expected to hold its next major event on September 9. That’s the natural stage for announcing new Apple TV and HomePod hardware — and for finally explaining why those devices were so quiet during the earlier software reveals.

What This Means for Your Smart Home

If Apple follows through, the living room becomes a much more capable place for AI. A smarter HomePod could handle multi-step requests, understand context better, and respond faster. An Apple TV with Apple Intelligence could turn your TV into a conversational hub — ask it to pull up a movie by mood, genre, or even a vague description like “that one with the guy from The Office.”

That kind of experience is already possible on phones. Extending it to the home is a logical next step. It also makes Apple’s ecosystem story stronger: one Siri, everywhere you are.

For now, it’s still speculation. But the beta evidence is hard to ignore. Apple is clearly testing the waters for Apple Intelligence on home devices. Whether it arrives this fall or later, the days of a dumb speaker in the corner may finally be numbered.