Apple’s new AI photo editing tools mostly work, for better and worse

At a glance:

• iOS 27 beta adds three AI‑powered editing tools: Clean Up 2.0, Extend, and Spatial Reframing
• Clean Up now runs hybrid on‑device and cloud models, fixing the artifact‑prone version from last year
• Spatial Reframing can shift perspective but may produce uncanny results, especially with close‑up subjects

What the new tools do

Apple’s iOS 27 developer beta introduces three AI‑driven editing features that fundamentally change what the native Photos app can accomplish. Clean Up 2.0 upgrades the earlier object‑removal tool by offloading part of the work to more powerful cloud‑based models, eliminating the blurry artifacts that made the on‑device version feel half‑baked. Users can now erase photobombers, stray objects, or even a booger from a child’s nose with a single tap, and the background is filled in convincingly.

Extend works like reverse cropping: it lets you add canvas around a photo and fills the new space with AI‑generated content that matches the existing scene. The algorithm prefers symmetrical extensions and avoids adding people, so it typically pads a landscape or adds missing elements such as a side‑mirror on a rally car. The amount of added area is limited, which curtails over‑ambitious “photoshop‑like” manipulation.

Spatial Reframing takes the concept a step further by simulating a slight camera movement after the shot was taken. The AI re‑composes the image as if the photographer had shifted position, giving a modest 3‑D‑ish effect. The tool respects the original perspective range—roughly the distance an arm could move—but can produce odd artifacts when the subject is close, sometimes distorting faces or inventing people who were never there.

How the tools compare to competitors

Google’s Pixel phones have been offering Magic Editor‑style capabilities for a few generations, relying heavily on cloud inference to produce seamless fills. Apple’s Clean Up 2.0 finally reaches a comparable level by adopting the same hybrid approach, while its earlier on‑device‑only version lagged behind with noticeable seams. Samsung’s early AI expansion experiments were more aggressive, often hallucinating entire objects; Apple’s Extend is deliberately conservative, adding only modest padding and avoiding people altogether. Spatial Reframing is unique to Apple’s ecosystem, building on the company’s existing depth‑map and portrait‑mode data, but its limited range means it can feel less polished than Google’s generative fill tools when handling complex foregrounds.

Practical implications for iPhone users

For the average iPhone photographer, Clean Up 2.0 will likely become the go‑to quick fix for unwanted background elements. Because the feature now leverages cloud models, it works reliably across a wide range of lighting conditions and textures, reducing the need for third‑party editing apps. Extend offers a subtle way to improve composition without re‑shooting, especially useful for landscape or product shots where a little extra breathing room can make a difference.

Spatial Reframing, however, introduces a new risk: the AI may subtly alter the scene in ways that feel “off,” particularly with close‑ups or selfies. The generated content can drift into the uncanny valley, making faces appear slightly skewed. Apple mitigates this by attaching a Synth ID label to any image edited with these tools, allowing platforms like Instagram to surface an “AI Info” badge. While the label is not a foolproof provenance guarantee, it signals to viewers that the image has been algorithmically altered.

Broader impact on photo authenticity

The rollout of AI editing directly inside the default Photos app lowers the barrier for anyone to manipulate images, potentially eroding trust in visual media. Even modest changes—such as adding a potted plant on a side table or moving an executive’s position in a group shot—can seed doubt about what was really captured. Apple’s decision to embed provenance metadata is a step toward transparency, but the effectiveness of such labels depends on how widely platforms surface them and whether users actively check the information.

What to watch next

Apple has not confirmed a public release date for iOS 27, and the tools remain in beta, meaning further refinements are expected before general availability. Key areas to monitor include improvements to the cloud inference pipeline for Clean Up, expanded padding limits for Extend, and more robust depth‑aware algorithms for Spatial Reframing that reduce facial distortion. As competitors continue to push generative editing forward, Apple’s conservative yet integrated approach may set a new baseline for on‑device AI photo workflows.