Apple is adding generative AI editing to the Photos app in iOS 27, but with guardrails that look deliberately tighter than what Google and Samsung ship. Two new tools, Extend and Spatial Reframe, generate fake pixels to expand a frame or shift its perspective, while an improved Clean Up tool erases unwanted objects with better infill. Extend expands a photo by 25 percent and can only be run once on a given image. The features were unveiled at Apple's Worldwide Developer Conference on June 11, 2026, and ship later this year.
Jon McCormack, Apple's iPhone camera chief, frames the rollout as targeted rather than expansive. He argues the tools exist to fix specific compositional mistakes — a plastic bag drifting through the background, a subject framed too close to the edge — rather than to let users rewrite a scene.
The constraints are unusually explicit for a consumer AI feature. The Photos models will not alter pixels on the main subject's face, and Clean Up cannot remove the primary subject of a photo. Extend cannot be chained: once you expand a shot 25 percent, you cannot save it and run Extend again to keep going. Apple is also tying repeat use of Extend, Spatial Reframe, and Clean Up to an iCloud subscription, though it has not disclosed the daily limits.
Key facts
- 01iOS 27 ships later this year with two new generative Photos tools: Extend, which expands a shot by 25 percent, and Spatial Reframe, which alters perspective.
- 02Extend works only once per image and cannot be re-run to infinitely enlarge a photo; Clean Up, Extend, and Spatial Reframe all carry daily usage limits tied to iCloud subscriptions.
- 03Apple will integrate Google DeepMind's SynthID watermark later this year so platforms can flag edited images as AI-altered.
- 04The Photos models are restricted from altering pixels on the main subject's face, and Clean Up cannot remove the primary subject of an image.
- 05Siri's Visual Intelligence is moving into the Camera app in iOS 27, but Siri cannot drive the new generative edits — those remain manual.
Apple plans to integrate Google DeepMind's SynthID watermarking technology later this year, embedding an invisible signal that downstream platforms can read to flag an image as AI-edited. Researchers have shown digital watermarks can be stripped or spoofed, but Apple is opting in to a cross-industry signal rather than building a proprietary one.
Della Huff, product manager for Apple's Camera and Photos software, says the models were tuned to minimize hallucination during edits. The training objective, she says, is to add as little new content as possible while still completing the user's request. In practice that means Extend will not invent a parked car beyond the frame of a street scene just because one is plausible.
The restraint is not absolute. In a developer beta test of Extend on a photo of a friend at a table, the tool added new tables and fake people in the background to match the aesthetic of the original. Huff says the model is trying to preserve the look of the scene: if the background has people in it, leaving the expanded area empty would read as wrong. "If we said the rule is we could never generate a background human ever, then the feature would become less useful," she said.
That trade-off is the entire debate over computational photography in one sentence. Google's Pixel team has leaned into the idea that a photo should match what the user remembers — bluer skies, better light, recomposed groups. Apple is staking out a narrower position: the Photos app is for repairing the shot you took, and Image Playground, Apple's separate text-to-image app, is where users go to generate something new. In iOS 27, Image Playground defaults to more photorealistic output unless the user picks an art style.
Siri is also moving into the Camera app in iOS 27, absorbing the Visual Intelligence feature that previously launched from the Camera Control button. McCormack says the move is about reducing friction — the camera is already a note-taking and lookup device, so the assistant belongs there. Siri can handle minor touch-ups in Photos but cannot drive Extend or Spatial Reframe; Apple's view is that perspective changes are too open-ended to express in natural language, which is the opposite bet from Google Photos, which added natural-language editing last year.
The skeptical read is that Apple is shipping a smaller feature set than its competitors and dressing the gap as principle. Extend's 25 percent cap, the single-pass restriction, and the subject-protection rules will frustrate users who want the kind of freeform recomposition that Google and Samsung already offer. Watermarks remain bypassable, and the line between fixing a frame and fabricating one gets blurry the moment the model invents background humans to fill out an aesthetic.
For Apple, the bet is that consumers want their camera roll to remain mostly trustworthy and their generative play kept in a different app. That is a defensible product position in a market where Google Photos, Samsung Gallery, and a long tail of App Store editors are racing the other direction. It also gives Apple a clean story for regulators and platforms now that SynthID is in the pipeline. The product question is whether the restraint reads as taste or as a feature deficit once iOS 27 reaches users this fall.
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.
