AI Photo Style Transfer in Mobile Apps: Implementation

We integrate AI-driven photo style transfer into mobile applications. The primary challenge for clients: Neural Style Transfer (NST) on mobile must not exhibit perceptible lag. The VGG-19 model has a footprint of 500+ MB, and processing a 512×512 frame on an iPhone 12 without hardware acceleration requires 3–4 seconds—unacceptable for user experience. Our accumulated expertise demonstrates that selecting the appropriate architecture is paramount. On-device inference is 10 times faster than server-side processing, which is critical for interactive UX.

The server path (leveraging Replicate, Stability AI, or a custom PyTorch backend) offers straightforward implementation, but latency typically ranges from 3 to 15 seconds. Conversely, on-device execution via CoreML or TFLite is more challenging but provides instant previews. Below is a comparative analysis of the two approaches.

In practice, we advocate a hybrid approach: on-device for quick previews (256×256) and server for final 4K export. This reduces server infrastructure costs by $2,000 per month for mid-scale apps, potentially saving up to $2,000 per month. Choosing the architecture is a critical design decision. Our engineers provide a consultation starting at $1,500 to evaluate your specific constraints.

How to Prepare a Model for Mobile Deployment?

Direct conversion from a PyTorch checkpoint to Core ML is not straightforward. The preparation involves four steps:

1. Train or obtain a Fast NST model in PyTorch (using torchvision.models or a custom architecture).
2. Export to ONNX via torch.onnx.export.
3. Convert using Core ML Tools: coremltools.convert(onnx_model, compute_precision=ct.precision.FLOAT16) → .mlpackage.
4. Validate on a device using MLModel.prediction(from:).

FLOAT16 quantization halves the model footprint without noticeable quality degradation. INT8 quantization is more aggressive but may introduce artifacts on texture regions. For Android, use tf.lite.TFLiteConverter.from_keras_model() → .tflite with post-training quantization.

Integration in iOS

import CoreML
import Vision

class StyleTransferProcessor {
    private let model: VNCoreMLModel

    init() throws {
        let mlModel = try FastNST(configuration: MLModelConfiguration()).model
        model = try VNCoreMLModel(for: mlModel)
    }

    func process(image: CGImage, completion: @escaping (CGImage?) -> Void) {
        let request = VNCoreMLRequest(model: model) { req, _ in
            guard let obs = req.results?.first as? VNPixelBufferObservation else {
                completion(nil); return
            }
            let ciImage = CIImage(cvPixelBuffer: obs.pixelBuffer)
            completion(CIContext().createCGImage(ciImage, from: ciImage.extent))
        }
        request.imageCropAndScaleOption = .scaleFill
        try? VNImageRequestHandler(cgImage: image).perform([request])
    }
}

Metal Performance Shaders are automatically utilized via the Neural Engine—no custom shaders required.

How to Manage Memory and Battery?

A common pitfall is keeping the model loaded persistently. On devices with 3 GB RAM (iPhone SE 2, budget Android), this triggers Jetsam kills. Best practice: initialize MLModel lazily on first use and unload after 10 minutes of inactivity. Ensuring stability is our commitment.

Battery: NST loads the Neural Engine. For live preview, we limit the frame rate to 10–15 fps using CADisplayLink with preferredFramesPerSecond. Full 30 fps on iPhone 14 Pro is possible but increases power consumption by 30%. User studies indicate a preference for balanced performance.

Handling Large Resolutions

CoreML models require a fixed input shape. If the model is trained on 512×512 and the user uploads a 48 MP photo, you must downscale before inference and upscale the result. Simple UIImage resize loses detail. We employ Guided Upsample via MPSImageBilinearScale or Real-ESRGAN for upscaling to 4K, ensuring high final quality.

Model Comparison for On-device NST

Fast NST with MobileNet is the optimal choice for mobile devices.

Server Path via Replicate

If on-device processing is not suitable, the Replicate API provides access to NST models. Example request:

POST https://api.replicate.com/v1/predictions
Authorization: Token <key>
{
  "version": "<model_version_hash>",
  "input": {
    "content_image": "<base64_or_url>",
    "style_image": "<base64_or_url>",
    "output_image_size": 1024
  }
}

Poll status every 2 seconds. Result arrives in 8–20 seconds. Always store the API key on a backend proxy—never on the client.

Included Deliverables

• Analysis and prototype: architecture selection (on-device / server / hybrid), model selection, performance estimation.
• Model preparation: training / distillation, quantization (PTQ/QAT), conversion to CoreML / TFLite.
• App integration: SwiftUI / Jetpack Compose, live preview, gesture handling, saving.
• Server side (if needed): FastAPI backend, Replicate integration, caching.
• Testing: on real devices (iPhone SE 2, 12, 14 Pro, Pixel 6, Samsung S22) with fps and power consumption measurements.
• Documentation and training: deliver source code, CI/CD, instructions for client's team.

With over 5 years of experience in mobile AI and more than 20 shipped projects, our team ensures robust delivery. We implement the feature turnkey.

Timelines and Cost

On-device integration of a ready model — 3–5 days (starting from $3,000). Full cycle including model selection, quantization, live preview, and server export — 2–4 weeks (estimated $8,000–$12,000). Cost is calculated individually after detailed requirements analysis. Our certifications and 5+ years of mobile development experience guarantee results. Contact us for a project assessment.

Neural Style Transfer

Machine Learning in Mobile Apps: CoreML, TFLite, and On-Device Models

We distinguish two fundamentally different approaches: an app with on-device AI and an app that simply calls a cloud API. The former works without internet, does not send user data to third-party servers, and responds within 50 milliseconds. The latter depends on network latency and pricing plans. Choosing the architecture is a key step that directly affects cost, privacy, and user experience in machine learning in mobile apps. Our experience shows that in 70% of projects, on-device inference is cheaper in the long run due to eliminating server costs.

How to Choose Between CoreML and TFLite for On-Device Inference?

CoreML — Apple's native framework for running ML models on device. Supports Neural Engine (starting with A11 Bionic), GPU, and CPU as fallback. Models are converted to .mlmodel format via coremltools from PyTorch, ONNX, or TensorFlow. Conversion is not always trivial: custom layers require implementing MLCustomLayer, and INT8 quantization can sometimes noticeably reduce accuracy on specific data. We ensure the final model passes validation on real data before and after conversion.

TensorFlow Lite — cross-platform alternative for Android and Flutter. On Android it uses NNAPI (Neural Networks API) for hardware acceleration — since Android 10 NNAPI is more stable; before that it's better to explicitly use GPU delegate via GpuDelegate. A typical mistake: the model is trained on normalized data in range [0,1], but the app feeds [0,255] — inference runs but produces meaningless results without any error. We include an automatic input data validation module in the SDK.

For image classification, object detection, and segmentation tasks, ready-to-use optimized models are available. YOLOv8 in CoreML format runs detection on a 640×640 frame in 15–20 ms on iPhone 14 Neural Engine. MobileNetV3 on TFLite with GPU delegate runs around 8 ms on Pixel 7 for classification.

What If You Need Text Generation On-Device?

Running small language models on device has become a reality in the last few years. Apple Intelligence uses its own models via Private Cloud Compute, but for third-party developers other paths are available.

llama.cpp with Metal backend on iOS is a working approach for phi-3-mini (3.8B parameters, 4-bit quantization, ~2.3 GB). Inference: 15–25 tokens/second on iPhone 15 Pro. For integration in Swift, use the Swift Package llama.swift or a wrapper via C interface llama.h. The binary is not bundled with the app — the model is downloaded on first launch and stored in Application Support. Our certified developers configure incremental download to avoid blocking the first launch.

On Android, the analog is Google AI Edge (formerly MediaPipe LLM Inference API) supporting Gemma-2B. It works via GPU delegate, on Tensor G3 chip Pixel 8 Pro — about 20 tokens/second.

Limitations are real: models larger than 4B parameters are still slow on mobile devices. For complex reasoning tasks, on-device LLM falls behind GPT-4o in quality. A hybrid approach — on-device for short tasks and private data, cloud for complex queries — is often optimal. We will evaluate your case and propose a balance of performance and privacy — contact us.

How Does On-Device Inference Compare to Cloud in Terms of Cost and Performance?

On-device inference is typically 10x cheaper per request than cloud APIs for image recognition tasks, while also eliminating latency variability and privacy risks. The table below summarizes the trade-offs.

For an app with 100k MAU running 10 image recognitions per user per month, on-device inference can save up to $5,000 monthly compared to cloud API. Get a free consultation on your ML architecture today.

Integrating OpenAI API and Other Cloud Models

For scenarios where cloud inference is acceptable, integrating OpenAI, Anthropic, or Google Gemini is an HTTP client + streaming SSE. In Swift, AsyncThrowingStream is convenient for streaming responses. In Kotlin, use Flow.

Critically: API keys must never be stored in the app bundle. Even an obfuscated key can be extracted from the IPA in 10 minutes using strings or frida. Correct architecture: mobile app → your own backend → OpenAI API. The backend controls rate limiting, logs requests, and protects the key.

What Is Included in the Work (Deliverables)

• Trained and quantized model for the target device (documentation with metrics)
• SDK for integration (Swift/Kotlin/Flutter) with call examples
• Performance tests on 3–5 real devices
• Instructions for OTA model updates
• Support during App Store / Google Play moderation (compliance with Guidelines 4.2, 5.1)
• 2 weeks of technical support after release

Typical Project Pipeline

1. Task analysis — measure latency, privacy, size, supported devices.
2. Model prototyping — in Python, evaluate accuracy on target data.
3. Conversion and quantization — for CoreML/TFLite with validation.
4. Integration into the app — model wrapped in a service layer (easy to swap CoreML ↔ TFLite ↔ cloud).
5. Testing — on real devices, measure FPS, RAM, battery.
6. Deployment — via TestFlight / Firebase App Distribution, monitor metrics.

Timelines: integration of a ready CoreML/TFLite model — 1–2 weeks, development of a custom model with mobile optimization — from 6 weeks, on-device LLM chat with personalization — 4–8 weeks.

Why We Take on Complex Cases?

10+ years of experience in mobile development, 50+ implemented AI/ML solutions, guarantee of compatibility with current iOS and Android versions. All projects undergo code review and load testing. The cost includes preparation of moderation documentation and training of your team.

Contact us — we will help you choose the architecture and implement ML in your app turnkey. Order an audit of your existing solution — we will assess the potential for server cost savings free of charge. In some projects, savings can reach significant amounts per month.