iOS 27 AI photo editing vs. Photoshop: mixed results in six-photo test

The comparison used six photos provided by the author’s friend Jürg, taken during travels in Switzerland and Europe. Three photos tested object removal and three tested image extension. For Clean Up, the author left the setting on Auto, one of three options alongside Fast and High Quality. For Photoshop, objects were selected with the Object Selection Tool or Lasso Tool, then Generative Fill was used with a blank prompt box.

For the Extend tool, the author used whatever extension Photos selected by default. According to 9to5Mac, Photos limits how far the edges of an image can be expanded, while Photoshop does not have the same restriction. To keep the comparison consistent, the author used Apple’s Photos result as a reference and had Photoshop create something in a similar size. The author also tried to keep HDR appearance consistent between the two editing apps, though minor variation remained.

In the first removal test, a red chair in front of a yellow door, Clean Up made a positive impression. The result looked believable; the only clear mistake was that a rivet behind the chair’s left armrest was removed rather than reconstructed. The tool replaced the shadow beneath the chair with a texture matching the surrounding scene while maintaining the existing lighting. Photoshop left the chair’s shadow in place, where it looked like an oil stain, misaligned the door’s rivets, left one obviously missing, and produced more visual artifacts. The author said the Photoshop result was not as clean or realistic as Clean Up’s.

In the second test, removing a plate with bread from a red table, Clean Up removed the plate and added a reflection of the mug and jars, including the handle of the spoon in the reflection. It also removed the plate’s shadow from the napkin. The author noted that removing a red plate from a red table of the same color might be harder for Clean Up, but it was not. Clean Up left crumbs from the plate, which the author said could be removed easily by running Clean Up again over that area. Photoshop removed the plate well enough, but left an out-of-place shadow and made the napkin look dirty or wet. It left fewer crumbs.

The third removal test was more difficult for Clean Up. After attempting to remove a man on the right, Clean Up left a garbled mess with obvious artifacts and leaves that were too chunky. It struggled similarly on the left side, where a woman with white hair stood behind white flowers, and it left a ghost of the dog’s head that blended somewhat with the stone wall but was more noticeable in a side-by-side comparison. Photoshop removed the woman behind the white flowers and the man behind the leaves without issues, realistically filling in both areas. The result looked convincing, with no one appearing to have ever been there, and Photoshop avoided leaving phantom shadows.

The comparison also included three image-extension tests, beginning with a photograph of old, winding steps. The author chose that photo because he thought neither AI would immediately recognize what it was looking at, and he was curious how each would interpret the scene. The article also noted that the author had used Photoshop’s generative AI features since Adobe introduced them in 2023, finding them hit or miss but faster than masking, cloning, and manually painting objects out.