Fast Style Transfer Flutter 🎨

A Flutter package for running Fast Neural Style Transfer using TensorFlow Lite (TFLite) models, enabling real-time image transformation on Android and iOS devices locally (no internet connection required).

💡 Based on the original Fast Style Transfer by Logan Engstrom and inspired by luiscib3r/style_transfer. Special thanks to them!

---

🧠 How It Works

1. Loads two models:
   • predict.tflite: extracts style features
   • transfer.tflite: applies style to the input image
2. Preprocesses the input content & style images
3. Runs both models sequentially
4. Returns JPEG-encoded image as Uint8List

---

📂 Supported Model Loading

You can load models from:

• Assets (recommended)
• File path (File)
• Raw bytes (Uint8List)

---

🚀 Installation

dependencies:
   fast_style_transfer_flutter: ^0.0.2

---

📸 Example

final styleTransfer = FastStyleTransferFlutter.init(
  config: FastStyleTransferConfig(
    thread: 4,
    useGPU: true,
    loaderConfig: FastStyleAssetsLoaderConfig(
      predictResource: 'models/predict.tflite',
      styleTransferResource: 'models/transfer.tflite',
    ),
  ),
);

final resultImage = await styleTransfer.run(
  request: RunTransferRequest(
    image: imageBytes, // content image (Uint8List)
    style: styleBytes, // style image (Uint8List)
  ),
);

// Use resultImage as Uint8List (e.g., display or save)

📸 Screenshots

---

📥 Model Download

You can use pre-trained models provided by TensorFlow:

• magenta_arbitrary-image-stylization-v1-256_int8_prediction_1.tflite
• magenta_arbitrary-image-stylization-v1-256_int8_transfer_1.tflite

These models are compatible with this package and ready to be used.

You can also train or convert your own models based on:

• Original: Fast Style Transfer
• Mobile conversion guide: luiscib3r/style_transfer

🙏 Credits

• Huge thanks to Logan Engstrom for the original Fast Style Transfer.
• Special appreciation to luiscib3r for mobile adaptation inspiration and open-source contribution.