It’s too early to draw firm conclusions about Apple‘s take on AI in the iOS 27 Dev beta, but I have been busy trying out new features, including the Apple Intelligence image-altering tools in Photos: specifically, Spatial Reframing and Extend.
I’ve already had some fun with both, and was immediately impressed with their raw power. To generatively alter your images, Apple uses a powerful private-compute cloud-based diffusion model built, in part, with Google. It’s unlike any AI Apple has ever presented before, and, as I’ve written, it opens Apple up to a lot of questions about whether it still prizes image truth over aesthetics.
I’m not here, though, to critique these tools. After all, this is the developer beta, and some tools and features will likely change quite a bit before they arrive fully baked on, we expect, the iPhone 18 in September.
Still, this platform update marks the beginning of Apple’s journey as a card-carrying AI citizen, fully capable of standing alongside OpenAI and partner Google. Siri can talk and pay attention, Image Playground can generate images from whole cloth, and Photos can alter and extend photos with a gesture.
Having spent time with Spatial Reframing and Extend, I remain somewhat startled at just how far Apple is taking the AI image-altering strategy. Spatial reframing, for instance, lets you turn photo subjects to see elements the camera never saw or captured.
Seeing beyond the edge
I’ve had more fun with Extend, though, not necessarily because I plan on filling in the lost information in my photos, but thanks to how Apple’s image-generation tools guess at what was never captured in the first place.
The thing about a tool like Extend is that, if you’re using your own photos, you already know what was cropped out. Perhaps you chose that framing, but as Apple noted in its keynote, it can be difficult to frame a portrait-mode photo in a landscape frame. Sometimes you need those extra bits.
Apple’s approach for the moment appears to be “let guesswork be your guide.” It doesn’t know what’s missing, but Apple Intelligence can use its smarts to read the rest of the image and fill in the blanks.
To see how this early version is doing, I took a bunch of photos of things and places, doing both a tight and a wide shot. In other words, I shot what was really there, and then cropped in to see just a smaller section. The goal: how close could Photo’s Extend tool come to filling in the blanks?
Considering what it had to work with, Photo Extend did a pretty good job, though a couple of the results were unintentionally comical.
One other thing I learned is that the Extend tool in the current version of iOS 27 Dev Beta will not extending, for instance, body parts. I took a photo of my hand, cropping out a few fingers, but no matter what I did, the app wouldn’t extend the frame to fill in the rest of my hand. This is actually good news; I had worries about ending up with six fingers.
I’m also learning that Extend likes order. It seems almost allergic to clutter, so whatever it does generate is usually clean, orderly, and has as few elements as possible.
Below are some examples of real wide shots alongside the more tightly framed ones, where I let Apple Intelligence Extend do its thing. If you hadn’t seen the originals, you might never know that significant portions of the images were generated by AI.
For the majority of these images, I took one photo for which I stepped back enough to capture more of the scene, then I took a second, tighter photo. I applied Extend to the latter image, and then compared Apple Intelligence’s guesswork to reality. You can see the original, unextended photo in the center.
Extending the frame and stretching reality
This is the only image where I took just one photo and extended it with Apple Intelligence’s latest Photo tools. This is a dev beta, so I won’t offer much criticism, but some might take issue with the Empire State Building redesign.
The home office
On the left, you can see my home office as it is. The orchid is a plant I’ve nurtured for years (I even fully replanted it almost 18 months ago). Extend’s gift is that it tries to leave original reality alone, but the more information you ask it to fill in, the wilder the flights of fancy.
I love, for instance, that my closet now has three doors, and I have been staring at the glass door on the right for hours. What is that? Why did Extend add it? No matter; it’s early days, and these generative skills are impressive.
Foliage
I think Extend already has a solid grasp of how plants and trees work, and shows off that knowledge here. The extended tree looks quite realistic. As for the sky, it now has a rather dramatic, almost beatific look. My shed got an odd redesign.
Guitar on the wall
The guitar work shows off a little of Extend’s penchant for cleanliness and order. The suggestion of a lampshade is turned into a perfectly round, brightly colored tube, and, based on the two visible supports, it might no longer be a lampshade.
The blinds also got a bit of a cleanup. As for my guitar, it looks essentially the same. In other words, Extend appears to be, even in Dev Beta form, ensuring that the image’s original subject remains — even if extended a bit — essentially untouched.
Office shelf
Something as busy as my office shelf can present challenges for any generative system — so many objects, so many options.
I took a picture (at left) of the actual shelf, and you can see the second, tighter photo in the center, and then the extended image on the right.
It makes sense that in the extended photo, Extend chose order over my clutter, but the best part is the transformation of the USS Enterprise into a quasi-jetliner. I think a couple of book titles also got a rewrite.
Backyard furniture
Extend did a nice job of decluttering my backyard space. I noticed that when Extend detects a cube-shaped object (in this case, my fire pit), it will just turn it into a featureless box, instead of trying too hard to guess.
Jetliner
I was impressed with how effectively Extend completed the jetliner. Notice, though, how once again it assumes order where, in truth, there’s chaos.
There you have it — some fun, early experiments with Apple Intelligence’s newest and most powerful image generation tools. That it does so well at the dev stage is especially encouraging, since Apple still has months to refine it.
Ultimately, I don’t think people will push Extend as far as I have. They may only use it to recenter a subject, adding just a few inches to one side or another, or to extend a solid background or even an evening sky, efforts that won’t detract from or alter the subject.
How do you think you’ll use these new Apple Intelligence Photo tools? Let me know in the comments below.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
https://cdn.mos.cms.futurecdn.net/c69b2o2M7gtgCjxR5fhsdL-2000-80.jpg
Source link
lance.ulanoff@futurenet.com (Lance Ulanoff)
