iOS 27 adds a suite of artificial‑intelligence tools to the native Photos app, including the Reframe, Extend and Clean Up editors, alongside a new Siri chatbot.
To evaluate these features, a series of images captured with an iPhone 17 Pro Max were processed, revealing a mix of impressive outcomes and notable shortcomings.
Reframe analyses a scene, then regenerates the photo from a shifted angle while restricting changes to areas the model understands; repeated adjustments are blocked by default but can be circumvented by saving and re‑importing the image.
When applied to a shot of Beirut’s corniche, Reframe accurately altered the handrail’s perspective and preserved the rainbow’s curve, yet it misplaced trash cans and streetlights.
A latte photographed in Byblos demonstrated Reframe’s strength: the first edit kept ice‑cube texture intact, and the second rendered the wooden table realistically, though the distant background appeared artificial.
Testing an airplane image exposed a failure mode: the AI bent the aircraft in the first edit and introduced unrelated elements in the second.
For a Ferris wheel at sunset, the initial Reframe attempt distorted the wheel’s circular shape, while a subsequent edit partially corrected the form but omitted several gondolas.
A lavender shrub overlooking a village yielded a highly realistic first edit with enhanced colors and clarity, but a second edit suffered from obvious hallucinations and repetitive patterns.
In a self‑portrait blowing a dandelion, Reframe’s first pass altered the stem and aged the subject’s appearance; the second pass further distorted the face yet correctly generated the lake surface.
Overall, Reframe can produce believable perspective shifts, yet inconsistent fidelity limits its practical usefulness for most users.
Extend expands an image by generating new content along its edges, employing similar safeguards against hallucination; the restriction can also be bypassed through repeated saves.
At the Port of Tripoli, attempts to extend a cropped lifebuoy and nearby boats were constrained to the left edge, resulting in underwhelming additions.
Conversely, extending a dimly lit car interior correctly identified the infotainment display in the first pass and accurately added the left mirror and window in the second.
Close‑up shots of cement boulders saw both extensions detect the rock patterns and produce realistic extensions.
Extending the base of a lighthouse successfully replicated sky colors and cloud textures; a subsequent extension added a lifelike sea but also hallucinated an extra, oversized lighthouse.
When extending buildings on Jounieh’s hillside, the AI filled the sides with generic, dull structures and inserted generic cars on an imagined road.
Extending a dark alley maintained the original atmosphere, generating appropriate cars and surrounding architecture.
These results illustrate Extend’s unpredictability: it may falter on well‑lit scenes while handling complex compositions effectively.
Clean Up now offers Fast and High Quality modes; Fast provides quick removal of obvious items, whereas High Quality analyses surrounding context for more realistic edits.
Removing pedestrians from a cobblestone street, the Fast mode produced a noticeably edited image, while High Quality yielded a much more natural appearance.
At the Port of Beirut, Fast removal of cranes and people left clear AI artifacts, whereas High Quality delivered a cleaner result, especially around textual elements.
Attempting to clear buildings obstructing a sunset view, Fast failed to produce a usable image, but High Quality generated a plausible scene despite subtle AI traces.
When tasked with erasing numerous objects in front of two kittens, Fast could not succeed, while High Quality accurately reconstructed the kittens and their paws.
Covering the photographer’s face tested the feature’s handling of human features; Fast created a distorted digital swirl, whereas High Quality generated a face that, while not a true likeness, was more coherent.
Cleaning a beach portrait, Fast struggled to remove items cleanly, while High Quality performed the edit professionally.
In summary, High Quality Clean Up consumes more processing time and power but consistently delivers reliable results, whereas Fast mode remains suitable for simple, quick blemish removal.
