Apple's AI Image Engine Is Getting a Serious Overhaul in iOS 27
Mark Gurman says Genmoji and Image Playground are in line for a 'big boost' in visual quality — and it's long overdue.
Written by OutOfToken AI
May 31, 2026 · 4 min read · Synthesized from reporting by 9to5Mac · How this works
Apple Intelligence launched with considerable fanfare, but its image generation tools — Genmoji and Image Playground — landed with a thud. The outputs were cartoonish, inconsistent, and miles behind what Midjourney, DALL-E, or even Google's Imagen were producing. Now, according to Bloomberg's Mark Gurman, Apple is preparing to course-correct in a significant way with iOS 27.
Gurman Drops the Intel
In his weekend Power On newsletter, Gurman — whose Apple sourcing is among the most reliable in the industry — reported that the image generation models underpinning Apple Intelligence are slated for a 'big boost' in visual quality ahead of iOS 27. The scope of the upgrade isn't fully detailed, but the implication is clear: Apple knows its current diffusion-based models aren't competitive, and engineering teams are working to close that gap before the next major platform release cycle. Gurman also flagged that Visual Intelligence — Apple's camera-based scene recognition system — is being considered for deeper integration directly into the Camera app, with increased customization options on the table.
The Problem With Apple's Current Image Models
Apple's on-device image generation pipeline, introduced with iOS 18, prioritizes privacy and local inference over raw output quality — a philosophically defensible position that has nonetheless produced mediocre results. Genmoji, designed to let users generate custom emoji from text prompts, frequently produces muddy, poorly proportioned figures that lack the crispness of even mid-tier web-based generators. Image Playground, the more ambitious feature, offers themed visual styles — Animation, Illustration, Sketch — but these guardrails have limited creative range and the outputs often feel generic. The constraint is partly architectural: running a capable generative model entirely on a Neural Engine, without server-side compute, demands aggressive compression and quantization that bleeds image fidelity.
"'Apple is preparing to substantially upgrade the visual quality of its image generation models for iOS 27' — Mark Gurman, Power On Newsletter"
What a 'Major' Upgrade Actually Requires
Closing the quality gap won't be trivial. State-of-the-art image generation today runs on transformer-based diffusion architectures with billions of parameters — models that demand GPU clusters to train and significant compute to inference. Apple could pursue several paths: a more aggressive hybrid model that offloads complex generation steps to Apple's Private Cloud Compute infrastructure while keeping prompt processing local; an architectural shift toward newer, more efficient diffusion transformer variants like DiT or flow-matching models that deliver higher fidelity at lower parameter counts; or simply raising the compute ceiling on what the A-series and M-series Neural Engines are allowed to consume during generation tasks. The upcoming A19 chip, expected in iPhone 17 Pro, reportedly features a substantially upgraded Neural Engine — which would give Apple the silicon headroom to run heavier models without torching battery life.
iOS 27 is still months from even a developer preview, and Apple's track record on AI features shipping as previewed has been uneven at best — Siri's long-promised contextual intelligence remains half-baked a year after announcement. But image generation is a battleground Apple cannot afford to concede. If Gurman's sources are right, and the visual overhaul is substantive rather than incremental, it could mark the moment Apple Intelligence finally earns its name in the domain users actually notice.
Editorial Note
The claim is attributed to Mark Gurman's Power On newsletter, which has a strong track record for Apple reporting, lending credibility to the source. However, iOS 27 is speculative (Apple typically announces major versions 1-2 years in advance), and specific claims about 'major visual upgrades' cannot be independently verified without official Apple confirmation. The assessment of current image quality being 'far from anything to write home about' is subjective opinion rather than verifiable fact.
Claim Tracker
AI-assessed
Subjective quality assessment; no objective benchmarks provided, though reflects widespread user sentiment from launch
Directly attributed to Bloomberg's Mark Gurman; specific source cited as weekend Power On newsletter
Core documented feature of Apple Intelligence since iOS 18 launch
Reported via Gurman but described as 'being considered' — indicates speculative/planned rather than confirmed feature
Ask AI about this story
// discussion
sign in to join the discussion