Apple has added three generative AI photo editing tools to the iPhone Photos app in the iOS 27 developer beta: a rebuilt Clean Up, a new Extend tool, and a Spatial Reframing feature that mimics moving the camera after the shot. The package, unveiled at WWDC 2026, is the first time the world's most-used camera app has shipped serious generative editing natively, and Apple is attaching Synth ID provenance labels to images touched by the tools.
The scope is narrower than what Google ships on Pixel phones with Magic Editor, and narrower than Samsung's earlier generative experiments. Apple has constrained how far each tool can push a photo, particularly around faces and the size of any extension. The features are still in developer beta and Apple may revise behavior before the public release.
“The most popular camera in the world just got its first set of serious AI photo editing features, and I don't think any of us are ready.”— Allison Johnson, Senior reviewer, The Verge
Clean Up is the most significant upgrade of the three. The previous version, introduced last year, ran purely on-device and frequently left artifacts where it tried to fill in details behind a removed object. The new Clean Up routes to larger cloud models when needed, the same architectural choice Google made years ago with Magic Editor, and the gap in output quality has closed. Removing a photobomber or a small distraction now produces convincing fill on typical iPhone shots.
Key facts
- 01iOS 27 developer beta adds three AI photo editing features to the iPhone Photos app: Clean Up, Extend, and Spatial Reframing.
- 02Clean Up now offloads to cloud models instead of running purely on-device, matching the approach Google has used for years on Pixel.
- 03Spatial Reframing limits perspective shifts to roughly the range of an arm's movement when the photo was taken.
- 04Edits are tagged with Synth ID provenance labels, which Instagram surfaces only inside an 'AI Info' submenu.
- 05The features were unveiled at WWDC 2026 and remain subject to changes before public release.
Extend is generative outpainting, framed as reverse cropping. Users can push the edges of a photo outward and let the model paint plausible content into the new margin. Apple has put hard limits on the tool: it refuses some directions outright, caps how much padding it will add, and appears to avoid extending across human subjects. In practice it leans on symmetry, completing a side mirror on a rally car or matching architectural lines, but it will also invent props — a potted plant on a side table in one test — that were never in the original scene.
Spatial Reframing is the most ambitious feature and the one most likely to cause trouble. It builds on the existing 3D-ish depth effect in Photos and lets users shift perspective as if they had physically stepped to one side when taking the shot. The adjustment range is roughly the span of an arm's movement at capture time. For distant subjects the edits are subtle and largely realistic. For close subjects, including selfies, the model has to synthesize more of the face and the results slip into uncanny territory quickly, with skewed features and invented detail.
The Verge's Allison Johnson tested Spatial Reframing on a photo from a WWDC tech talk and found the model invented a person sitting next to Craig Federighi when the framing shifted enough to reveal a previously obscured area. The tool's two-dimensional cousin, Extend, performs better because predicting plausible pixels in a flat plane is a simpler problem than reconstructing a three-dimensional scene from one viewpoint.
Apple is tagging edited images with Synth ID, the content-provenance standard developed by Google. Instagram reads the metadata when users upload affected photos, but the platform only surfaces the AI disclosure inside a secondary 'AI Info' menu rather than on the image itself. Provenance labels in general remain easy to strip with a screenshot or a re-export, and the iOS 27 implementation does nothing to change that underlying weakness.
“But even introducing a little bit of doubt about the provenance of a houseplant on a side table, or whether someone was indeed standing exactly where they appear to have taken a photo from, can add up to a lot of trouble over time.”— Allison Johnson, Senior reviewer, The Verge
The competitive context matters. Google has had cloud-assisted generative editing on Pixel for multiple generations, and Samsung shipped its own version with more permissive defaults. By comparison, iOS 27's editing suite is restrained: limited extension distances, face-avoidance in Extend, and a Spatial Reframing range capped at arm's length. Apple has chosen the conservative end of the design space, and the company will still be the largest distributor of generative photo editing on the planet by a wide margin the moment iOS 27 ships.
The concern that lingers across the feature set is not whether any single edit looks convincing. It is that the iPhone has historically been treated as a relatively trustworthy source of casual documentary images — group photos, news from a phone in the room, evidence in a dispute. Once the default camera app can routinely insert plausible objects, slide a subject's position, or invent a face that was never there, the baseline assumption that an iPhone photo reflects what the lens saw weakens. Synth ID labels help in theory and do almost nothing in the social-media flows where most photos actually travel.
For Apple, the trade-off is straightforward product math. Clean Up alone will be one of the most-used features in Photos within a quarter of release, because removing strangers and small distractions is a near-universal want. Extend will pull in a smaller but loyal audience of users who reframe for Instagram. Spatial Reframing is the wildcard — useful when it works on distant subjects, and a liability the moment a high-profile manipulated image with a Synth ID label still goes viral. Apple's restraint buys it time, but the company has now joined the same generative-editing race it spent years avoiding, and the next iOS will be judged on whether the guardrails hold.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
