iOS 27 Photos AI Features Explained: Spatial Reframing, Extend, and Apple's Guardrails

Two Apple Photos executives gave their clearest public account yet of what the iOS 27 Photos AI features are built to do. The word they kept returning to was "refine." The flagship tool, Spatial Reframing, can shift the apparent camera angle of a photo after the fact and change where a subject appears to be looking. That's a harder sell for "refinement" than Apple's framing acknowledges.

In a new interview with Tyler Stalman, covered by 9to5Mac today, Photos executive Huff said the goal is to take a recorded memory and help users "refine" it, then drew a distinction between two senses of that word: cleaning up distractions versus adjusting composition or perspective. The guardrails Apple has built around these tools are substantive. AI generation fires only where a perspective shift creates missing pixels. The Extend tool is capped at 25% per side and restricted to a single application per image. Every AI-edited photo carries a hidden SynthID watermark identifying it as machine-modified. What the WWDC demo showed is that "refinement" can produce results that go further than the framing implies.

iOS 27 Spatial Reframing: how it works and what the demo revealed

Spatial Reframing lets users shift the apparent camera position of a photo in post-processing, adjusting angle, perspective, and composition after the shot is taken, according to Apple. The feature draws on depth map data captured at the moment of shooting, the same spatial information the iPhone uses for Portrait mode, to build a 3D model of the scene. Where a perspective shift opens visual gaps, AI fills them; where it doesn't, the original image data stays untouched, MacRumors explained this week.

For images without native depth data, the tool can generate a depth map algorithmically, per PCMag. That matters because part of Apple's credibility argument rests on edits being grounded in what the camera actually captured. An algorithmically derived depth map is an approximation, not a record.

Huff was explicit in today's interview that generation only fires where the perspective shift demands it. The feature is designed to stay "true to the original moment," with new content introduced only where the shift makes it unavoidable, 9to5Mac reported.

The WWDC demo told a more complicated story. Apple engineer Alok Deshpande showed a photo of his children in the front yard before school, slightly off-angle relative to the lens and too centered for his taste. He used Spatial Reframing to shift the angle and adjust the crop. The result looked as if it had been taken from a lower perspective, as though he had dropped to one knee, and his daughter appeared to be looking more directly into the lens, PCMag reported from the keynote.

Changing where a subject appears to be looking is not a compositional fix. It's an alteration to what actually happened in front of the lens. Apple's position, quoted by 9to5Mac today, is that "there was nothing that became fake about it" but the demo suggests the tools can produce results that go well beyond tidying a frame.

Extend and Clean Up: the stronger case for Apple's "bounded editing" argument

Extend and Clean Up are easier to defend on Apple's own terms. Both change what surrounds or intrudes on the subject rather than the subject itself, which keeps them closer to the refinement end of the spectrum.

Extend pushes out the borders of an image to add visual breathing room, straighten a crooked horizon, or fit a different aspect ratio, filling the new space with generatively produced background content, Apple explains. Photos executive McCormack said in today's interview that the 25% cap per side came from internal testing, and the single-use restriction exists to prevent compounding drift away from the source image.

Clean Up gets upgraded AI models in iOS 27, now capable of handling larger and more complex removals with more natural-looking infill, according to Apple. Huff described the updated feature as able to "clean up more complex objects and handle bigger tasks while still staying true to the original moment."

Together, these two tools show what Apple's guardrails look like when they hold cleanly: hard caps, one-use restrictions, changes confined to background and peripheral elements. Spatial Reframing shares the same design principles but operates closer to the subject. The results, as the demo showed, can affect the people in the frame in ways that are harder to wave away.

What the guardrails cover and what they don't

The constraints Apple has built are real. Generation is confined to gap-filling, expansion is hard-capped, edits are single-use, and source-image geometry sets the floor for what the AI can produce, per 9to5Mac and Apple's own documentation.

The provenance layer is also real. Every AI-edited photo in iOS 27 will carry a hidden SynthID watermark identifying it as machine-modified. What Apple hasn't addressed publicly is how that watermark gets surfaced for anyone receiving these photos on social platforms, in group chats, or in professional contexts. Hidden metadata is only useful if something can read it.

That gap matters most when the use context isn't the family album. Removing a distracting trash can from a vacation photo and altering the apparent gaze of a person in a news image are not equivalent acts, even if both happen inside the same Apple Intelligence tools menu. The features themselves don't enforce that distinction. The user does.

For everyday photo sharing, these tools are likely useful and low-stakes. For anyone whose images carry evidentiary weight, documentary photographers, journalists, legal records, the question of what "refine" means will need a more complete answer than Apple has offered so far.

What to expect in practice

The tools are restricted to Apple Intelligence-capable hardware, starting with the iPhone 15 Pro, which makes them simultaneously a product feature and a premium-device differentiator. Apple chose Photos, the most emotionally loaded app on the phone, as the venue to make that argument.

Extend and Clean Up sit comfortably within Apple's "refinement" framing and will serve most everyday editing needs without much friction. Spatial Reframing is more capable and more complicated. It's genuinely useful for fixing a crooked horizon or opening up a tight group shot. It's worth pausing before applying to any image where the accuracy of what actually happened matters.

The real test isn't the interview or the keynote. It's how these tools perform on older photos, low-light images, and scenes without native depth data, and whether the SynthID watermark ever becomes readable to anyone other than the person who made the edit.