Apple’s latest AI photo editing features show mixed results.

The world’s most popular camera just received its first major AI photo editing features, and it’s safe to say we might not be fully prepared for the impact. While the new capabilities in iOS 27 may seem modest compared to what’s available on devices like Google’s Pixel phones, they mark a significant shift in the iPhone’s native photo editing capabilities. It’s all about enhancing those precious memories, right?

Currently, these features are part of the iOS 27 developer beta, so keep in mind that there may be further adjustments before they roll out to everyone. There are three new AI editing tools in this update, though one of them, Clean Up, feels more like a half-feature since it was previously available but not very effective. This tool allows you to remove unwanted elements from your photos, and it has received a substantial upgrade this year. The other two features are Extend, which lets you expand the edges of your photo using AI, and Spatial Reframing, which simulates moving the camera around to recompose an existing shot. This last one is the most ambitious and potentially the most problematic.

Let’s start with Clean Up. It’s actually effective now. Instead of relying solely on on-device models to remove objects and fill in details, it can now tap into more powerful cloud-based models. This is a game-changer, as it brings the tool up to par with Google’s Magic Editor, which has been superior for years. The previous version of Clean Up struggled to convincingly replace removed elements, often leaving behind odd artifacts. Clean Up 2.0? It gets the job done.

Using AI to edit photos is something I’m comfortable with. I’d use it to remove a distraction from a family photo or eliminate an unwanted stranger from the background. The new version in iOS handles these tasks seamlessly, and I anticipate it will be a hit among iPhone users.

Next up is Extend, which is like cropping but in reverse. It allows you to stretch the edges of your frame, which can be useful if your composition feels too cramped. However, it has its limits; it avoids altering people in the photo and may restrict the direction in which you can extend. It adds just a bit of padding, which helps prevent misuse. Like Clean Up, it performs convincingly, often seeking symmetry in its edits. For instance, it added a part of a rally car that was cut off in my original shot, seamlessly including a side mirror to match the existing one.

While it doesn’t seem overly eager to fabricate elements like some earlier attempts from other brands, I did notice it added a potted plant to a side table in one instance. It looked fairly convincing, but I’d hesitate to share that on social media knowing it wasn’t real.

Then there’s Spatial Reframing, which introduces a third dimension to your photos. This feature builds on an existing capability that gives your images a 3D effect, allowing you to reframe a shot as if you had physically moved the camera. You can only adjust it to a certain extent—about as far as you could have reached when taking the photo. This is perfect for those moments when you realize you could have improved the framing by shifting slightly to the left to avoid distractions.

However, this feature can lead to some unexpected results. I tested it on a photo taken at a tech talk, where one of the executives was mostly obscured. When I adjusted the framing, the AI generated an extra person sitting next to Craig Federighi, which was a bit unsettling.

The challenge of predicting what should appear in a three-dimensional space is evident in the results from Spatial Reframing compared to Extend. The outcomes can be odd, especially when the subject is further away, limiting your ability to recompose effectively. In close-up shots, the adjustments can be more dramatic, but the AI has to work harder to fill in gaps, which can lead to uncanny results. Faces can appear skewed or distorted, and the AI may invent details that weren’t there. While the idea of rescuing a poorly composed photo is appealing, the execution can leave much to be desired.

A small consolation is that images edited with these AI tools receive Synth ID labels indicating they’ve been modified. When I uploaded a couple of these images, Instagram recognized the edits, but you have to dig into the “AI Info” menu to see that information. These labels aren’t foolproof, and the bigger concern is the diminishing trust in photos shared online. While Apple isn’t the most radical player in this space, even a little uncertainty about the authenticity of a photo can lead to significant issues over time.