Google Pixel’s AI features ranked

Google's Pixel lineup has long trumpeted its AI and machine learning capabilities as a unique selling proposition, and this year's Pixel 9 stable is no exception. With the release of the Pixel 9 lineup, Google has pulled some new AI rabbits from its hat, and the company has been pretty vocal about how cutting edge and practical they are.

But, AI is a phrase that gets bandied about in the tech world so much that it's lost all meaning. It's a marketing buzzword that gets attached to products and pushed by PRs, with only vague explanations of what anyone would actually use or need the features for.

So, with that in mind, I took it upon myself to test drive all the new AI-powered features of the new Pixel lineup. I wanted to see not only if they were useful, but if they actually worked.

I needed some sort of metric by which to judge all the AI features, and I figured I could combine my two great passions: snarky commentary and the Cartesian coordinate system. That's why I created this handy X-Y axis to graph each Pixel feature:

On the X axis, I'll be ranking the features from a scale of "Cool Party Trick" to "Undeniably Practical". Sure, some AI features are neat, but do they actually have a practical use case? We've been sold the idea of AI as a massive leap forward for productivity, but so far, I've lost more productivity pleading with ChatGPT to rewrite The Dukes of Hazzard as a Lovecraftian body horror than I've ever gained. That's why it's important to look at these features not just for what they can do, but why we would need them to do it.

On the Y-axis, I'll be ranking each feature on a scale of "Works as Advertised" to "Janky Mess". The most practical AI feature in the world is pretty useless if it doesn't actually do what it says on the tin.

Now that we all understand the metrics, let's dive into the features and see where they fall on my authoritative and not-at-all hastily slapped together in Canva graph.

Let's start off with a softball. Pixel Studio is Google's answer to services like Dall-E. It's an image generator that allows you to type a prompt, and spits out an AI picture.

When you first open Pixel Studio, you're hit with a disclaimer that it can't generate images of humans ... yet. It does say that this feature is coming in the future, but doesn't specify a timeframe.

This immediately threw me off my game. I had intended to give it the most obvious prompt: a shirtless Mikhail Gorbachev wrestling Waluigi. But it was not to be. Instead, I decided on my fallback option: a donkey riding a hovercraft. This is what it returned:

By God, that's a donkey. On a hovercraft. No one can tell you it isn't. Pixel Studio nailed that brief. It's obvious that this feature works pretty well as advertised. I even used it to generate the featured image for this article. I have to rank this as "Works as Advertised".

But what about its practicality? You could make a strong argument that art, in and of itself, is utilitarian, as it fulfils a basic human need to create and to observe the aesthetically pleasing. But I find the idea that AI-generated images could ever be considered art ridiculous and offensive. So I have to relegate this to "Cool Party Trick". I probably would've ranked it higher on the practicality scale if it could have shown me the former Chairman of the Supreme Soviet grappling with Nintendo's most underrated villain, but c'est la vie.

Next I turned my attention to the new AI feature in Google's notes app. It promises to be able to help you make lists or organise your thoughts in note form. Pretty handy.

It gave me a few suggestions for how I might use it, including a shopping list for a party of 20 people, things to do on a weekend in New Orleans or the best ways to wind down after a long day of work. I decided to go the more esoteric route, and asked it for a list of the best weird fiction novels. This is what it gave me:

Colour me impressed. I've read a lot of these, and they're definitely the books I'd recommend to someone interested in weird fiction. It also gave me some good recommendations I haven't read. All in all, it did a great job, and not just because it gave me smug validation. This feature definitely Works as Advertised.

I can also see it being highly practical. Producing shopping lists or brainstorming ideas would be much easier and more efficient with its help. I'm ranking this one high on the Undeniably Practical scale.

Who wants to remember things the way they happened when you can remember them the way you wish they happened? That's the premise of Google's AI photo features. One of the methods you can use to Total Recall your precious memories is Reimagine. This lets you select an area or object in the photo, type in a prompt and produce an AI-enhanced copy.

I chose literally the first photo I came across in my camera roll, because I am nothing if not a lazy, lazy man. It was a sunset shot of the carpark outside the skateboarding school my daughter attends. Here's the original:

When you enter the photo editor on your Pixel and tap the Magic Editor button, Reimagine will pop up as a suggestion. Tap an area of the photo to highlight it, and a text box will pop up, allowing you to type in a prompt. I selected the car in the foreground and typed "red". Sure enough, the car turned red.

Next, Reimagine suggested I do something about that uninspiring sky. The majesty of the natural world is a bit "meh", after all. Why not juice it up? I took Reimagine's advice, and was treated to a sky painted in glorious hues of red and orange. Google: 1, Nature: 0.

Finally, I decided to get rid of the unsightly carpark. I simply typed "jungle", and lo and behold, I was given a red car in the jungle at sunset. How it got there is a question for the more cynical among us. I simply basked in its beauty. Here it is:

Reimagine gave me exactly what I asked for. But, the picture is quite clearly AI-generated. It just has that uncanny valley look about it, so I can't give it a perfect Works as Advertised score, but it's pretty good.

As for practicality, it's mostly a parlour trick. But I can see it having some use cases for people who lie on Instagram or in helping real estate agents misrepresent properties without resorting to poorly-executed Photoshops. So I'll put it smack in the middle between Cool Party Trick and Undeniably Practical.

This might be the feature closest to my heart. Like most people, I take a lot of screenshots. One has to keep one's meme game strong in these uncertain times. But, when the occasion calls for it, where to find one Simpsons meme among hundreds? Enter Screenshots.

If you opt into AI Experiences when you're setting up your Pixel, your phone will automatically categorise and label your screenshots. It'll begin processing them the first time you open the Screenshots app, and will apply labels to them that make them searchable.

I found this worked a treat. Say I wanted to track down the time I accidentally solved Wordle in two? No problem. I typed Wordle into the search bar, and up it popped.

Or, what if I wanted to share the screenshot demonstrating how hilariously bad Google's AI search summaries are? You know, the one that told me that Andrew Jackson, who was famously from Tennessee and died in 1845, graduated from the University of Wisconsin at Madison in 2005? Easy. I just started typing "president", and before I could even finish, my screenshot appeared.

Screenshots is a great feature, AND it works. Best of all, it works in the background to categorise your screenshots. It's high on the Works as Advertised scale.

I also put it fairly high on the Undeniably Practical scale. Sure, its primary use case might be tracking down memes, but I defy you to rubbish the practicality of pulling out just the right meme at just the right time.

Oh, Gemini. Google really hyped you. Sure, no one asked for you. We were getting along fine with Google Assistant. But Google decided we needed a smarter, more intuitive virtual assistant to help us complete our daily tasks.

I was actually pretty optimistic about Gemini. With its paid Gemini Live service, Google said you could have a free-flowing conversation with the AI assistant. You definitely can. The only problem is Gemini is not very bright.

For my first test, I didn't want to tax the poor AI too much. I decided to go with one of its own suggestions: quiz me about classic sci-fi movies. So Gemini started throwing out plot synopses and asked me to guess the film.

Which '80s classic comes to mind when you hear this? A group of kids must band together to save their small town from an alien invasion. Why, it's The Goonies, of course! What's that you say? The Goonies features nary a single alien, let alone an entire invasion? Congratulations. You know more about film than Gemini.

What about this one? In this classic neo-noir, in a city constantly under attack from flying cars, a detective must solve a murder.

It's' Bladerunner, silly goose! What? Rick Deckard wasn't under attack from flying cars? Too bad. I actually want to see that movie.

I also asked Gemini to tell me how many times the letter "z" appears in the word "pizza", and in spite of having to TYPE OUT THE WORD to answer my question, it confidently informed me that the letter "z" does not appear anywhere in the word "pizza".

Even more frustrating, each time I corrected Gemini, it would respond with "You're absolutely right! My apologies". Don't patronise me, Gemini.

All of this is pretty discouraging, given Gemini has the potential to be the Pixel's most practical feature. It's also the one Google is positioning as a paid service, through its Gemini Advanced. You can use Gemini without paying (and Gemini Advanced is free for a year when you buy an eligible Pixel device), but the Gemini Live features, along with AI features for Google Docs and Sheets, are paywalled. I can't for the life of me imagine forking over $32.99 per month to have a robot lie to me.

While Gemini ranks highly on the Undeniably Practical scale, I have to rate its current iteration as Janky Mess. Maybe it'll get smarter in time, but early adopters beware.

Zoom Enhance is another handy photo feature that zooms in on a photo and uses generative AI to fill in the pixels. Considering most of the Pixel lineup already has pretty impressive telephoto abilities, this could seem like a hat on a hat.

In practice, though, it works remarkably well. Here's a photo I originally took at 5x zoom, albeit with the Pixel 9 Pro Fold's comparatively underpowered telephoto lens:

It's already a wee bit pixelated, so I didn't hold out a lot of hope for Zoom Enhance. Imagine my surprise, then, when this is the result I got:

Holy cow, that's better than the original. There's a bit of blurring around the margins, and the text on the tag in the upper right corner isn't readable, but all in all it's pretty impressive.

Of course, this is working with an object with well defined edges and largely symmetrical features. It might not perform as well on, say, a human. But it certainly did enough with this photo to impress me. It rates highly on the Works as Advertised scale.

As for practicality, it's not entirely without its practical use cases. It's more of a novelty than something I'd use every day, but I can definitely get some good photo edits out of it. I rate it as just sneaking into the Undeniably Practical range.

Add Me is both one of the coolest and most troubling photo features in the new Pixel range. It allows you to add yourself to group photos. You snap a shot of your friends, then switch places with them and nestle in close to an AR afterimage while someone snaps your photo.

It's all a bit alarming. First, there's the distortion of reality. Then, there's the weird social dynamic that could develop where you're always the one in the friend group taking the group shot, never the one in it. You've basically been relegated to the role of a sentient tripod.

Now, testing this feature would require me to have a group of people around. But since I wrote this article while I was home by myself, I had to be content with a sleeping cat and a stuffed wombat. Here's the result:

I hear what you're saying: "But, Adam, that's just a picture of a toy wombat next to a cat. Big whoop!" Well, what if I told you that when I took the photo of the wombat, the cat WASN'T EVEN THERE?

It ... it was. But in a slightly different position. TECHNOLOGY!

So, the feature worked, albeit with less-than-ideal testing conditions. I'll have to defer to my colleague, Dylan Crismale, who actually has a social life, and said in his Pixel 9 review that he got inaccurate results using the Add Me feature. I'm going to rate this about midrange on the Works as Advertised scale.

As for practicality, this is more cool novelty than handy feature. It's more practical than Pixel Studio, but perhaps even less practical than Reimagine. I'm going to rate it between them on the Cool Party Trick scale.

By and large, the Pixel lineup's new AI features work very well. Surprisingly, most of them are pretty utilitarian. It's just a shame that the most utilitarian feature, Gemini, is also the most broken. But overall, one has to hand it to Google. Inspecting that final graph, there are a lot more dots in the 1st quadrant than in the 4th. Also, it looks a bit like the graphed function of a fourth degree polynomial. WHAT ARE WE TO MAKE OF THAT? I'd ask Gemini, but I wouldn't trust the answer.

But in the areas where people expect Pixel to excel, photo editing and image processing, its AI features are largely kicking goals. Add onto that some genuinely handy productivity hacks, and I'd say the Pixel lineup has acquitted itself well. Now if you'll excuse me, I'm going to work on my screenplay about Harrison Ford being attacked by flying cars.