← Back to blog
AiAbout 6 min read

Mainland China iPhones Finally Get Apple Intelligence: Image Generation Hidden Behind Three Unassuming Buttons

Published Nov 7, 2026
Mainland China iPhones Finally Get Apple Intelligence: Image Generation Hidden Behind Three Unassuming Buttons

In the early hours of November 4, Apple pushed iOS 27.3 Developer Beta 1, and for the first time mainland China iPhones can turn on “Apple Intelligence & Siri” in Settings. Mainland China users have waited more than two years since the feature first appeared at WWDC in June 2024.

According to Apple, the official release will roll out to all Chinese users with iOS 27.3 in December. The first feature list is not long: Live Translation, Writing Tools, Visual Intelligence, Clean Up, Genmoji, natural-language photo search, Memory Movies, priority email, and email summaries.

There is no item on the list called “AI image generation.” But if you take these features apart, you will find that image generation has already quietly entered four of the system’s most everyday places.

Qwen Handles the Part You Can’t Come Up With

On October 31, Apple and Alibaba announced a partnership. Qwen has been integrated into Apple Intelligence across three areas—Siri, Writing Tools, and Visual Intelligence—providing text and image understanding as well as content generation capabilities.

The division of labor is roughly this: Apple’s own generative models handle the system-level basics. Live Translation, rewriting and proofreading, Clean Up, and Genmoji all run on Apple’s own models. Requests that need longer context, more world knowledge, or connections to external services such as Taobao are handed to Qwen.

In practice, the boundary is clearer than the description. Ask Siri “I haven’t been sleeping well lately,” and it calls Qwen to give advice on routines and environment. If you want to go deeper, tap the Qwen icon to jump into the app, where the conversation history syncs automatically and you do not have to repeat yourself. If you want to send down jackets to your parents in Heilongjiang and ask whether they can arrive in three days, Siri gives the delivery time; further price checks and booking a door pickup are completed by jumping to Qwen and invoking the SF Express agent. If you like a guest’s outfit on a variety show, first ask Siri for the brand, then jump to Qwen to find the same item on Taobao.

In these steps, image capabilities appear in two places: Visual Intelligence adds “Ask Qwen,” so you can point at an object or scene and directly identify it and ask questions; Writing Tools adds “Qwen Create,” which can generate content from scratch. Alibaba’s announcement explicitly mentions “including images.”

Three Buttons, Three Different Things

The first is Clean Up in the Photos app. This is handled by Apple’s own model, and the official line is that it “removes distracting objects while preserving the scene exactly as it was captured.” It sounds like the object removal every phone has, but the difference is that it handles light and shadow relationships more conservatively. The goal is not to leave an abrupt clean patch after erasing something.

Overhead view of blank white photo cards on a warm wooden desk, one card lifted with a soft eraser beside it.

The second is Genmoji. Enter a sentence of description, and it generates a custom emoji. The technical difficulty of this has been underestimated: it has to compress semantics into a recognizable image on an extremely small canvas, and it also has to align with the strokes of the system font. Apple has made it a system-level capability that does not require opening any third-party app.

The third is natural-language photo search. Type “that photo from the beach last year with the dog,” and it pulls the image directly from your photo library. This feature looks like retrieval, but what it actually tests is image understanding. When there are tens of thousands of photos in the library, the accuracy of the tagging system determines the experience, and the tags are generated by the model.

Add Memory Movies, which generate a short film with transitions from a description, and the system’s image capabilities already cover four things: generation, editing, retrieval, and orchestration.

The Entry Point Changed, and So Did the Business

Over the past two years, domestic AI image generation products have competed over model scores and price. 7B open-source models can run on consumer-grade GPUs, an image costs a few cents, and a poster costs 0.3 yuan. These advances are real.

The problem is that users have to open a webpage, install an app, register an account, and think up a prompt. Every step loses people.

What Apple has done this time is put image capabilities into three actions that require no decision: tap once while cleaning up photos, type a sentence while wanting to send an emoji, type a line while looking for a photo. Users do not need to know which model is behind it, nor do they need to know what a prompt is.

For domestic vendors, this is both a threat and an interface. The threat is that Apple will take away part of the defaults at the system-level entry point; the interface is that Qwen, as a partner for Apple Intelligence in mainland China, pushes “Chinese large models” to the system layer of top global consumer hardware. Wang Xiaojie of Beijing University of Posts and Telecommunications judged in an interview that this is the first time a top global consumer hardware manufacturer has used a Chinese large model as the main backbone for system-level AI in the Chinese market.

The privacy design deserves a separate mention. Apple says it will obtain consent before sending any question, document, or image to Qwen; users can choose “Sign in with Apple” or use their own Qwen account; Alibaba explicitly states that it will not use user requests or generated results to train models. Whether such commitments can be verified is another matter, but writing them into the product flow is a precondition for China-market AI features to launch.

The Parts That Haven’t Arrived

Before the official December release, several things are still undecided.

Device restrictions are strict: iPhone 15 Pro / Pro Max, iPhone 16 and later, iPhone Air, and iPhone Duo. Device language and Siri language must be set to Chinese or English. iPad and Mac are awaiting approval.

Feature restrictions also remain: Live Translation supports a limited number of languages, Visual Intelligence can recognize a limited range of categories, and image generation itself is not a canvas for free play—it is embedded in specific forms such as emoji, Memory Movies, and writing illustrations.

The real change may not be on the feature list. Over the past two years, the adoption path for AI image generation in China has been to acquire users through tools: whichever company has a stronger model and lower prices, users run there. Now there is another path: capabilities follow system updates, and users do not need to be persuaded—they only need to tap that button on some afternoon.

This path may not be comfortable for model companies, and it is even less comfortable for app companies. But for ordinary people who need to clean up photos, send messages, and write invitations every day, having one less app to install is a tangible benefit.

Related articles