With a myriad of Gemini-powered AI features, the Pixel 9 is practically a “Gemini phone” as MKBHD mentions in his review. While Google is pushing for more on-device AI, most of these AI features still require an internet connection. So when you tap that button for some AI magic, data is sent to and from Google’s cloud servers where the heavy lifting happens. It’s like outsourcing the AI tasks instead of solely relying on the Tensor G4 chip.
So, do Pixel 9 AI features need an internet connection? Generally, yes. But do all of them? No. I’ll explain.
Table of Contents
Pixel AI Features That Need You to Be Online
- Gemini AI Assistant, Pixel Studio, Video Boost, Night Sight Video, Magic Editor
For starters, the Gemini AI Assistant requires an internet connection to work. Yes, unlike Apple’s Siri or even the Google Assistant, where you can do a lot of things offline, neither the Gemini Assistant nor its Gemini Live feature will work without internet connectivity.
I’m hoping Google will eventually allow it to do at least a few things offline, like opening apps, turning on the flashlight, or toggling your phone settings, but that’s not currently possible.
With the Pixel 9, Google introduced a new Pixel Studio app that uses AI to create images from text prompts. While it utilizes on-device processing, it also relies on Google’s cloud-based Imagen 3 for generating images, thus requiring an internet connection. The app simply won’t work without it.
Apple AI’s image generator, on the other hand, will be completely on-device. That makes it much faster and privacy-friendly compared to Google’s implementation.
Should you choose to use the new Video Boost feature to fix shaky, grainy, and otherwise imperfect videos, it would require the video to be uploaded to Google Cloud and processed, only after which you’ll be able to watch the boosted video. It consumes a lot of data and takes some time.
Night Sight Video on Pixel Pro phones is powered by the same feature, you’ll need to be online for that as well.
Again, I hate to break it to you, but the Magic Editor in Google Photos requires an internet connection to work. So every time you want to edit your photos, they’ll be doing a handshake with Google Cloud, which also means there won’t be instant edits.
Pixel 9 AI Features That Don’t Need the Cloud
- Pixel Screenshots, Zoom Enhancer, Add Me, Call Notes, Magic Eraser, Audio Magic Eraser
Pixel 9 comes with a brand-new Screenshots app that scans all your screenshots, understands what’s in them, and makes it easy to organize and recall information. For example, the other day I took a screenshot of Geekbench benchmarks I ran on the Pixel 9, and it cleverly fetched and prepared a note with all the scores. This feature works completely on-device.
Beyond screenshot management, the Pixel 9 boasts several other impressive on-device AI features like the Magic Eraser which lets you remove unwanted objects or people from photos, seamlessly filling in the space. The popular Add Me feature too doesn’t need it for obvious reasons.
The new Zoom Enhancer feature can also work offline, enhancing details in photos when you pinch to zoom in. Think of it as Pixel’s Super Res Zoom, but it works after you’ve taken the photo. Audio Magic Eraser, which removes unwanted noise from videos, is another existing on-device feature.
Finally, with Call Notes, your Pixel 9 can now automatically record any calls on your device and then show you a transcript and summary of the conversation seconds after you end the call. This works completely on-device, and no data leaves your device.
On-Device vs. Cloud Processing: Why it Matters?
- On-device: Faster, private, limited capabilities.
- Cloud: Slower, less private, but vastly more powerful.
With on-device processing, all your AI computation and processing are done within the device. No data leaves your phone. Since there’s no need for data transfer to and from the cloud, tasks like generating or editing photos/videos using AI run much faster. It’s also more privacy-friendly and works everywhere, even when you don’t have Wi-Fi or mobile data.
However, AI requires a lot of computational power, and the quality of its output is limited by how powerful the device is. This is why the new Apple Intelligence is only available on the iPhone 15 Pros, powered by the A17 Pro chipset, and not on lower-end models.
Cloud processing means you virtually have access to unlimited computational power. Resources can be easily scaled up when needed, and AI models can be updated and improved more frequently without the need for device-level updates.
However, your data is sent to the cloud for processing and then downloaded back afterward. This can be concerning for your private data. It’ll also consume a lot of your internet bandwidth and will work incredibly slowly when you don’t have high-speed connectivity. The features are simply unusable offline.
Why Connect to the Internet After All?
As was the case with the Pixel 8 and 8 Pro, most of the AI features released this year with the Pixel 9 and 9 Pro don’t actually run completely on-device, although that has been a term highlighted in bits and pieces. I see no straightforward clarity from Google directly on which features work entirely offline and which ones don’t.
That, I feel, is partly because the new Tensor G4 is a smaller upgrade than what everyone presumed it to be. It’s more or less a Tensor G3 with one core less, slightly better performance and efficiency. Google even used the same Tensor Processing Unit as last year, with enhanced Gemini Nano being the only new thing here.
Considering the Tensor G4 isn’t inherently as powerful or even close in processing power as its counterparts like the Snapdragon 8 Gen 3, Apple A17 Pro, or even the Exynos 2400, it’s clever of Google to offload most of the load to its cloud servers.
That also means these cloud-based AI features can later be passed onto older Pixel phones. And just like how Google made Magic Eraser, Photo Unblur, and Magic Editor free and available to all Android and iOS users, these may fall in the same ballpark. I wouldn’t be surprised if Google include them as part of the Google One subscription benefits.
Google is a cloud and software company more than it is a hardware company, after all, and this approach suits its direction better. But I wish to see more completely on-device AI features as the whole ‘AI on phone’ thing becomes more commonplace.
