📸 InstaMeme — On-Device AI Meme Generator for iOS

InstaMeme is a fully on-device, privacy-first meme creation app powered by Apple Silicon and the MLX framework.
With just a photo and a tap, InstaMeme generates short, punchy meme captions using local inference—no internet connection required.

Users can:

• Take a live photo or pick from their library
• Auto-generate meme captions using Vision + Llama running in MLX
• Remix memes endlessly
• Share or export them as burned-in images
• Build and extend a persistent personal meme gallery

InstaMeme transforms casual camera moments into share-ready memes instantly—all processed locally on ARM hardware.

---

🚀 InstaMeme Highlights

InstaMeme demonstrates on-device AI with a fun user experience.

This project showcases:

Category	Achievement
Technological Implementation	Uses Apple’s MLX, on-device LLM inference, Vision classification, SwiftData persistence, and zero-server architecture.
User Experience	Designed as a frictionless meme workflow: snap → suggest → remix → share.
Impact	Demonstrates a pattern for developers to incorporate on-device LLMs and generative UX flows in consumer apps.
WOW Factor	Watching a live camera image turn instantly into a meme—with no network.

InstaMeme showcases how the next wave of mobile apps can use fast, private, localized intelligence powered by Arm.

---

✨ Core Features

• 🧠 On-Device Caption Generation
  Uses MLX + a quantized Llama-based language model for offline meme text.

• 📷 Live Camera Capture OR Photo Library Import
  Built with modern SwiftUI + PhotosPicker integration.

• 👀 Apple Vision Recognition
  Vision framework detects objects in the image, feeding semantic context into the LLM.

• 🖼️ Burn-In Graphics Rendering
  Exported memes are rendered into a final flattened image, guaranteeing compatibility with SMS, Messenger, WhatsApp, etc.

• ♻️ Remix Mode
  Users can regenerate new captions or edit them manually, creating variations without losing prior versions.

• 💾 SwiftData Storage
  Fully persistent offline gallery that never requires a connection.

• 🔄 Resizable Image Pipeline
  Automatically scales images to mobile-friendly dimensions to prevent memory overuse when sharing.

---

🛠️ Tech Stack

Component	Technology
UI	SwiftUI
Storage	SwiftData
ML Framework	MLX + MLXLLM
Computer Vision	Apple Vision Framework
Model Execution	Quantized on-device Llama variant
Rendering	UIKit + SwiftUI ImageRenderer, Core Graphics burn-in
Platform	iOS (Apple Silicon / A-Series ARM chips)

---

📦 Setup & Build Instructions

Requirements

• macOS Tahoe or later
• Xcode 26 or later
• Apple Silicon Mac
• iPhone running iOS 17+ (required for live camera + on-device ML performance)

1️⃣ Clone the Repository

git clone https://github.com/MkFoster/InstaMeme
cd InstaMeme

2️⃣ Open in Xcode

open InstaMeme.xcodeproj

3️⃣ Add Required MLX Packages

In Xcode:

File → Add Packages…

Add these URLs:

• https://github.com/ml-explore/mlx-swift.git
• https://github.com/ml-explore/mlx-swift-lm.git
• https://github.com/ml-explore/mlx-swift-llm.git

Confirm they are attached to the app target in:

Project Settings → Target → Frameworks, Libraries, and Embedded Content

4️⃣ Build and Run on a Physical Device

Because the app uses:

• Camera input
• Vision framework
• MLX running on-device
• Neural Engine acceleration

…it must run on real hardware.

Select your device in Xcode and press ⌘ + R

5️⃣ First-Run Model Download

When you tap Suggest Captions for the first time:

• The app will automatically download the quantized MLX model.
• It is stored locally and reused — no internet needed afterward.