Apple explains why its approach to AI photo editing is different from rivals: Accuracy over fantasy

Trending 1 month ago

Serving tech enthusiasts for complete 25 years.
TechSpot intends tech study and proposal you can trust.

A basking potato: One of nan galore questions being raised astir AI-packed smartphones relates to their expertise to edit photos. At what point, for example, does an image go truthful altered that it nary longer qualifies arsenic a photograph? It's thing Apple is good alert of, which is why Cupertino wants its editing instrumentality to connection a much realistic consequence than those of its rivals.

Apple's suite of artificial intelligence tools, aptly called Apple Intelligence, yet arrives connected compatible devices pinch nan motorboat of nan iOS 18.1 update connected October 28.

One of nan caller features of Apple Intelligence is Clean Up, a instrumentality successful nan Photos app that tin region group and objects from images.

Apple's Senior Vice President of Software Engineering, Craig Federighi, sat down pinch The Wall Street Journal's Joanna Stern for an question and reply astir Apple Intelligence. At 1 point, nan newsman demos Clean Up, showing really objects specified arsenic a h2o vessel tin beryllium edited retired of a photo.

Stern past notes really immoderate of Apple's rivals specified arsenic Samsung and Google connection much extended AI devices that make parts of photos. She uses nan summation of a lion and an detonation created by Google Reimagine arsenic an example. Apple doesn't let users to adhd their ain contented to images.

When asked astir Apple's contrasting approach, Federighi's bottommost statement was "We thief purvey meticulous information, not fantasy."

"Do we want to make it easy to region that h2o bottle, aliases that mic? Because that h2o vessel was location erstwhile you took nan photo," Federighi said.

"The request for group to want to cleanable up what look for illustration extraneous specifications to nan photograph that don't fundamentally alteration nan meaning of what happened has been very very high, truthful we've been consenting to return that mini step."

The advancing tech down generative AI is making heavy edited and deepfaked images harder to place arsenic fakes. It's partially why Apple went successful a different guidance to its rivals. "Even nan expertise to region that h2o bottle, location were tons of debates internally," Federighi said.

"We are concerned that there's a awesome history to photography and really group position photographic contented arsenic thing they tin trust connected is suggestive of reality."

Even Apple's Image Playground, which allows users to create images utilizing concepts for illustration themes, costumes, accessories, and places, has definite limitations.

"When you look astatine experiences for illustration Image Playground, we made judge that nan images we were generating were not photorealistic," Federighi said. "Not because nan underlying exemplary couldn't make thing photorealistic but because we ne'er wanted personification to person immoderate disorder of whether Joanna was really wearing that fuzzy hat?"

Some companies offering AI devices place images arsenic being altered successful nan metadata. Apple does it pinch those edited utilizing Clean Up, and it besides tags them successful nan Photos app arsenic "Modified pinch Clean Up."

Further highlighting nan quality successful attack betwixt nan 2 companies, Samsung responded to disapproval of nan Galaxy S24's AI photo-editing features successful February by saying location is nary specified point arsenic a existent picture.

More
Source Tech Spot
Tech Spot