AI-Generated Product Descriptions in Mobile Apps

How does AI generate product descriptions from photos?

A marketplace seller takes a product photo with their phone and presses 'Publish'. The problem: manual description takes up to 30 minutes, quality and consistency suffer. We solve this with a combination of Vision API and LLM — the system offers pre-written text based on the photo and product category in 2–4 seconds. Our team has over 10 years of experience in mobile development and has completed 40+ AI service integrations. Time savings for the average store — up to 40% of a manager's workday, and copywriting costs are reduced by 70%. One medium-sized store saves $2,500 monthly on copywriting costs. Our basic integration costs $2,000 and typical ROI is achieved within 3 months.

How to implement AI product description generation in a mobile app?

Visual analysis of photos is the first step. Google Cloud Vision API returns object tags, color, brands on packaging, text on labels (OCR). For mobile apps, a multimodal LLM (GPT-4o, Gemini Pro Vision) is more convenient — a single request with an image analyzes and generates text at once. Object recognition accuracy reaches 95%.

Structured attributes from the form — the user fills in a minimum: category, price, condition (new/used). This data is included in the prompt as structured context. The rest the model infers from the photo.

Client-Side Implementation

The entire flow is asynchronous: the user selects a photo, presses 'Create description', sees a skeleton loader, and receives editable text in 2–3 seconds.

class DescriptionGeneratorViewModel : ViewModel() {

    fun generateDescription(imageUri: Uri, category: String) {
        _uiState.value = UiState.Loading
        viewModelScope.launch {
            try {
                val base64Image = imageUri.toBase64(contentResolver)
                val response = descriptionApi.generate(
                    GenerationRequest(
                        imageBase64 = base64Image,
                        category = category,
                        language = Locale.getDefault().language,
                        maxLength = 300
                    )
                )
                _uiState.value = UiState.Success(response.description)
            } catch (e: Exception) {
                _uiState.value = UiState.Error(e.message)
            }
        }
    }
}

On iOS similarly using async/await + URLSession:

func generateDescription(image: UIImage, category: String) async throws -> String {
    let imageData = image.jpegData(compressionQuality: 0.8)!
    let base64 = imageData.base64EncodedString()

    let request = DescriptionRequest(imageBase64: base64, category: category, language: Locale.current.languageCode ?? "ru")
    let response = try await api.generateDescription(request)
    return response.text
}

The image is compressed to JPEG quality 0.8 before sending — this reduces payload size from ~3 MB (RAW from camera) to ~300–500 KB without noticeable quality loss for Vision API.

Prompt Engineering for Quality Results

def build_prompt(category: str, image_tags: list, language: str) -> str:
    return f"""
You are a professional copywriter for an online marketplace.
Write a product description based on the provided image.
Category: {category}
Detected attributes: {', '.join(image_tags)}
Language: {language}

Requirements:
- 2-3 sentences, 50-100 words
- Start with the main product feature, not "This is a..."
- Include detected color, condition, and brand if visible
- Use active voice
- No adjectives like "great", "amazing", "perfect"
"""

The ban on 'great', 'amazing', and 'perfect' is not a formality. Models by default insert them into every other sentence, making descriptions indistinguishable.

Streaming for Improved Perceived Performance

Instead of waiting for the full response, use streaming via Server-Sent Events. Text appears as it is generated, like in ChatGPT. On Android, this is implemented using okhttp3.EventSource; on iOS, using URLSessionDataTask with the didReceive data delegate. This is especially important for long descriptions (100+ words) so the user doesn't wait 4–5 seconds in emptiness.

Handling Low-Quality Descriptions

Implement a hybrid approach: AI generates a draft, a human finalizes it. According to client feedback, this method reduces manual work by 60%. For automated quality control, use A/B testing of two description versions — AI vs. AI+editor — and track conversion.

Why Choose AI-Generated Product Descriptions?

Our AI-generated product descriptions are 5x faster than manual writing, improving efficiency by 80%.

Deliverables (What's Included)

• API and integration documentation
• Access to the repository with prompts and configs
• Team training on template editing
• Launch support
• Recommendations for A/B testing descriptions

Avoiding Common Integration Mistakes

• Sending RAW without compression increases response time. Always compress to JPEG quality 0.8.
• Ignoring the category — the model generates generic text. Always pass the category in the request.
• No fallback when Vision API fails — the user sees an empty screen. Always show a placeholder or error message.

Work Process

1. Analysis: define product categories and image sources.
2. API design: request format, Vision API error handling.
3. Configure prompt templates by product category.
4. Develop client UI with skeleton loader and result editor.
5. Implement streaming to improve UX for long descriptions.
6. Test on real products and A/B comparison.

Estimated Timelines

Basic integration (photo → description via GPT-4o / Gemini) — 3–4 days. With category-specific prompts and streaming — up to 1 week. Complex projects with on-device processing — from 2 weeks.

Order an integration estimate today — contact us. Get a consultation on integrating AI into your app. Contact us for a demo and project evaluation.

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.