Apple AI Photo Editing in iOS 27: New Tools, Tested

Apple AI photo editing has always lagged behind what Android users take for granted. With iOS 27, that gap is finally — and meaningfully — starting to close. Three new or significantly updated tools have landed in the iOS 27 developer beta, and they tell two very different stories: one about Apple catching up to the industry, and another about just how philosophically complicated it is to let AI rewrite your memories.

• Apple AI photo editing finally gets a major upgrade in iOS 27 with three new or improved tools across the native Photos app.
• The revamped Clean Up tool now taps cloud-based models, closing the gap with Google’s Magic Editor after years of falling short.
• Spatial Reframing is the most ambitious of the new features, but it can generate faces and objects that simply weren’t in the original scene.
• AI-edited images receive SynthID labels that platforms like Instagram can detect, though users must actively look for that information.

Clean Up Finally Does What It Promised

When Apple introduced the Clean Up tool in a previous iOS release, the reception was polite at best. The feature — designed to erase photobombers, distracting objects, or that unfortunate piece of litter in the corner of an otherwise perfect shot — leaned entirely on on-device machine learning models. The results were often blotchy, unconvincing, and riddled with the kind of artifacts that made you wish you’d just cropped the image instead. Most people quietly stopped using it.

iOS 27 changes the equation. Clean Up now offloads the heavy lifting to cloud-based models when needed, a move that immediately puts it in the same category as Google’s Pixel phones and their well-regarded Magic Editor suite. Google has been using server-side processing for generative photo edits for years, and the quality difference was always obvious. Apple held back, presumably prioritising its privacy-first, on-device processing ethos — but the results simply weren’t good enough.

The new version works. Removing a stranger from the background of a family photo, cleaning up a blemish, erasing a rogue road sign — Clean Up handles all of it without the uncanny smearing that plagued earlier attempts. It’s the kind of tool that will quietly become one of the most-used features on the iPhone for millions of people who’ll never once think about how it works. That’s a compliment, not a criticism. The best AI editing is invisible.

Of the three new capabilities, Clean Up is also the one that raises the fewest philosophical questions. You’re removing something real to restore an image closer to what you intended to capture. That feels defensible, even if nothing about the scene was technically how the camera saw it. As a core part of Apple AI photo editing, it sets a reassuringly high bar for the tools that follow.

Extend: Reverse Cropping, With Caveats

Extend is where Apple AI photo editing starts getting more interesting — and slightly more uncomfortable. The feature works like cropping in reverse: instead of cutting the frame down, you expand it outward, and the AI generates plausible-looking content to fill the new space. Got a portrait where your subject is a little too close to the edge? Extend gives them room to breathe.

The implementation is notably conservative, which is probably the right call for a first attempt. Extend won’t dramatically widen a shot or conjure entire new scenes. It adds modest padding — enough to fix a slightly cramped composition, not enough to fabricate an entirely different environment. Apple also appears to have built in guardrails around people: the tool seems reluctant to extend or alter human figures, steering the generation toward background elements instead.

In testing, the results were largely convincing. The tool filled in a portion of a rally car that had been clipped at the edge of the frame — including a matching side mirror — without the composition looking obviously artificial. It gravitates toward symmetry, which tends to produce believable outputs. It’s not flawless, though. A potted plant appeared on a side table in one expanded image — plausible enough that you might not question it, but wholly fabricated. That’s the kind of detail that would feel dishonest on social media, even if it’s technically harmless.

Samsung reportedly drew heavy criticism for its early generative editing features, which were enthusiastic about inventing dramatic new details — sometimes entire background elements — with little restraint. Apple AI photo editing, by contrast, is clearly playing a different game with Extend, one focused on credibility over spectacle. Whether that continues as the feature matures remains to be seen.

Apple AI Photo Editing Gets a Third Dimension — and It Shows

Spatial Reframing is the most technically ambitious of the three tools, and the most uneven. It builds on the depth-mapping tech Apple uses to give still photos a 3D parallax effect, extending it so you can recompose a shot as if you’d physically shifted position before pressing the shutter. Stepped a little too far to the right and caught a lamppost in front of your subject? In theory, Spatial Reframing lets you fix that.

The constraints are sensible on paper. You can’t dramatically relocate the camera — the adjustment range roughly maps to how far you could have physically moved your arm in the moment you took the shot. The goal is correction, not reinvention.

In practice, results vary wildly depending on the scene. Subjects that are far from the camera tend to produce relatively stable reframes, since the AI has less compositional work to do and the generated content stays subtle. But anything involving a closer subject — particularly a person — can deteriorate fast. Selfies are a notable weak point: changing the perspective means the AI has to reconstruct parts of your face it can’t actually see, and the outputs can veer into uncanny valley territory with faces that look subtly skewed or misshapen.

The more alarming failure mode is outright invention. In one documented test using a photo taken at an Apple event following the WWDC keynote, an attempt to shift the framing — originally needed because the photographer was seated off-angle — resulted in the AI generating what appeared to be a new person seated next to Craig Federighi onstage. Not a distorted reflection of someone real. A fabricated figure, placed credibly in the frame. That’s a different category of problem from adding a potted plant to a living room, and it’s a reminder that Apple AI photo editing carries real responsibilities at this level of ambition.

Three-dimensional scene reconstruction is a harder computational problem than two-dimensional inpainting, and the gap in output quality between Extend and Spatial Reframing reflects that clearly. Apple’s constraint on the adjustment range helps, but it doesn’t eliminate the risk of plausible-looking fiction being inserted into what’s supposed to be a document of something real.

The Bigger Question: What Is a Photo Now?

Apple isn’t the first company to wrestle with this, and it won’t be the last. But because the iPhone is the most widely used camera on the planet, the philosophical stakes of these decisions are unusually high. Google, Samsung, and Adobe have all introduced generative editing tools, with varying levels of transparency and restraint. What sets this moment apart is scale: when Apple ships something to iOS, it reaches over a billion active devices. These aren’t niche tools for power users — they’ll be default options in the app that most people have never opened settings in. Apple AI photo editing, deployed at that scale, shapes how billions of people understand what a photograph even means.

Apple has added SynthID metadata labels to AI-edited images, which is a meaningful step. Instagram reportedly reads this data and can flag affected images with an ‘AI Info’ label — though that information is buried behind a tap rather than displayed prominently. That’s a reasonable start, but it’s worth asking whether passive metadata is adequate when the average user has no idea what SynthID is or where to find it.

The tools are still in developer beta, and Apple routinely refines features between preview and public release. Clean Up, at least, seems ready. Extend is useful with appropriate expectations. Spatial Reframing feels like something that needed more time in the lab — technically impressive in narrow conditions, but prone to errors that are hard to spot and potentially easy to misuse.

The trajectory is clear, though. Apple AI photo editing is only going to get more capable and more integrated with each iOS cycle. The capabilities on offer now represent just the opening chapter of a much longer story about how Apple AI photo editing will reshape consumer photography. The real challenge isn’t technical — it’s figuring out where the line is between a useful correction tool and something that quietly turns photographs into something more like illustrations. That’s a question the industry hasn’t answered yet, and iOS 27 won’t be the last software update to force it back onto the table.

Source: The Verge

Frequently Asked Questions