Add the clothing model to Xcode

xcode-select --install
xcodebuild -version
# In Xcode: File > New > Project > App, Interface: SwiftUI
# Add CoreAI to the target and add the .aimodel to Copy Bundle Resources.
# Xcode Settings > Components > install Metal Toolchain.

Add and inspect the asset

1. Open an Xcode 27 iOS or macOS app target.
2. Drag clothing.aimodel into the project.
3. Make sure it is in the app's resources or model build phase shown by Xcode.
4. Open the Core AI model viewer. Check the function name, tensor names, shapes, and data types before you write Swift.

The portable file is .aimodel. A compiled, architecture-specific file uses .aimodelc.

Load and run

import CoreAI

let modelURL = Bundle.main.url(
    forResource: "clothing",
    withExtension: "aimodel"
)!

let model = try await AIModel(contentsOf: modelURL)
let function = try model.loadFunction(named: "main")

// Build this from 28x28 grayscale pixels in [0, 1].
// NDArray is float32, NCHW, with shape [1, 1, 28, 28].
let image: NDArray = makeImageNDArray(
    values: pixels,
    shape: [1, 1, 28, 28],
    dataType: .float32
)

let outputs = try function.run(
    inputs: ["image": image],
    states: [:],
    outputViews: nil
)
let logits = outputs["logits"]

main, image, and logits are the names from the conversion lesson. Confirm them in the model viewer. Build the NDArray from the real NDArrayDescriptor and initializer in the SDK you have installed.

Preprocess like training

For a picked photo, resize the CGImage to 28×28, draw it in grayscale, and scale its pixel values to 0–1. Fashion-MNIST pictures have a dark background and light clothing, so a normal photo may need a crop, contrast change, or inversion to look like the training examples. The next lesson shows the image path.

Load the model when the app starts

Load and specialise the model during app setup, not inside the photo picker callback. Specialise means get the model ready for this device. You can use the path below and keep the prepared model in memory so later photos are faster:

let model = try await AIModel.specialize(contentsOf: modelURL)
let function = try model.loadFunction(named: "main")

The exact method names can change. Check them against the Core AI documentation that ships with the Xcode 27 SDK.

Sources: Core AI documentation and WWDC26 sessions 324 and 325.