[Bug][iOS] SDXL (iOS) Core ML model fails to load because validator expects Unet.mlmodelc but Apple bundle ships chunked UNet files #217

Description

On iOS, after downloading and selecting the built-in SDXL (iOS) (Core ML) image model, the app fails to load it and shows:

Failed to load model: Missing required model component: Unet.mlmodelc

This appears to be a regression in the Core ML model validation logic, not a bad download or user workflow issue.

Steps to Reproduce

1. Install the iOS App Store build of Off Grid.
2. Open the image model picker.
3. Download/select SDXL (iOS) (Core ML).
4. Try to load the model.

Expected Behavior

The SDXL iOS Core ML model should load successfully.

Actual Behavior

The app rejects the model with:

Missing required model component: Unet.mlmodelc

Why this looks like a bug

The iOS validator currently requires these files for all Core ML models:

• TextEncoder.mlmodelc
• Unet.mlmodelc
• VAEDecoder.mlmodelc

and for SDXL also:

• TextEncoder2.mlmodelc

However, Apple’s current apple/coreml-stable-diffusion-xl-base-ios package exposes:

• TextEncoder.mlmodelc
• TextEncoder2.mlmodelc
• UnetChunk1.mlmodelc
• UnetChunk2.mlmodelc
• VAEDecoder.mlmodelc

So the app is rejecting a valid SDXL bundle because it assumes SDXL uses a single Unet.mlmodelc.

Likely Regression

This appears to have been introduced by the March 14, 2026 validation hardening commit:

• 834d78c – prevent CoreML TextEncoder crash in Stable Diffusion pipeline

That change added the pre-load file validation. Earlier SDXL support was added/fixed in:

• PR fix: SDXL model crash on iOS - use correct pipeline type #88 – use StableDiffusionXLPipeline for SDXL models

Suggested Fix

Update the iOS Core ML validator to accept SDXL layouts that use chunked UNet bundles, e.g. UnetChunk1.mlmodelc + UnetChunk2.mlmodelc, instead of requiring only Unet.mlmodelc.

It would also be good to add tests covering both:

• classic Core ML layouts with Unet.mlmodelc
• SDXL layouts with chunked UNet bundles

Environment

• Platform: iOS
• Install source: App Store
• Device: iPhone 17 Pro Max
• iOS version: 26.3.1
• App version: 0.0.81

Screenshots

Two screenshots are available showing:

• the SDXL (iOS) (Core ML) model selected in the image model picker
• the resulting error dialog with Missing required model component: Unet.mlmodelc