Make it Flow

Make it Flow is a 100% on-device iOS App & CLI tool that converts PDF documents into beautifully formatted, highly readable EPUB files. Powered by CoreML YOLO models and PDFKit, it intelligently reconstructs complex document architectures entirely offline.

🚀 Getting Started

Requirements

• macOS 14.0+ / Xcode 15.0+ / iOS 17.0+

Running the iOS App

1. Clone the repository: git clone https://github.com/codinguniversefromEric/Make_it_Flow.git
2. Open Flow_1.xcodeproj in Xcode.
3. Select a simulator or physical device, and press Cmd + R to build and run. (YOLO models are built-in, no external downloads required)

Running the CLI

Ideal for batch processing, automations, or CI pipelines.

cd Flow_CLI
swift run Flow_CLI /path/to/input.pdf /path/to/output.epub [nano|small|medium]

flowchart LR
    A([Terminal Command]) --> B{"Args Parser"}
    B -->|input.pdf| C["Load PDFKit"]
    B -->|Model Selection| D["Load CoreML Weights"]
    
    C --> E["Batch Processor"]
    D -.->|nano / small / medium| E
    
    E --> F["Make it Flow Core Engine"]
    F --> G["Generate EPUB / TOC"]
    G --> H[/"Save to output.epub"/]
    
    style A fill:#f96,stroke:#333,stroke-width:2px
    style H fill:#9f6,stroke:#333,stroke-width:2px
    style F fill:#69f,stroke:#333,stroke-width:2px,color:#fff

Loading

🧠 Architecture Overview

The system pipeline is designed for high-performance extraction of academic papers and heavily formatted books:

flowchart TD
    A[/"Input PDF"/] --> B["PDFKit & VisionEngine"]
    
    B -->|Rasterized Images| C["CoreML YOLO DLA model"]
    B -->|Native Rich Text| D["Hybrid Corrector"]
    
    C -->|14-Class Bounding Boxes| D
    
    D --> E["Layout Sorting & Stitching"]
    E --> F["EPUB Synthesizer"]
    
    F --> G[("Publisher-Grade EPUB")]
    
    style A fill:#f96,stroke:#333,stroke-width:2px
    style G fill:#9f6,stroke:#333,stroke-width:2px
    style C fill:#f9f,stroke:#333,stroke-width:2px
    style F fill:#69f,stroke:#333,stroke-width:2px,color:#fff

Loading

1. VisionEngine: Executes YOLO to classify 14 types of layout regions (Title, Body, Table, Picture, Formula, etc.).
2. Hybrid Corrector: Combines YOLO's skeletal bounding boxes with native coordinate data extracted by PDFKit to perform precise alignment and text assignment.
3. Layout Sorting: Computes geometric topological sorting, Y-axis alignment, and handles cross-page sentence stitching.
4. EPUB Synthesizer: Compiles the parsed fragments into a strict, publisher-grade EPUB ZIP archive with responsive CSS.

📊 Models & Benchmarks

The project comes pre-packaged with three lightweight CoreML models based on hantian/yolo-doclaynet. All models execute 100% offline leveraging Apple's Neural Engine.

Model	Average Processing Time (per page)	Layout Similarity (Accuracy)	Edit Distance	Recommended Use Case
YOLOv26 Nano	~0.05 seconds	~0.60	0.710	Fast, battery-efficient. Good for standard text documents.
YOLOv26 Small	~0.05 seconds	~0.65	0.656	Good balance of speed and precision for standard devices.
YOLOv26 Medium (Default)	~0.06 seconds	~0.68	0.612	Default. Highest accuracy. Ideal for complex, multi-column academic papers with dense tables and charts.

Note: Benchmarks were measured on Apple Silicon (M-series) against the Hugging Face marker_benchmark dataset. The built-in Swift engine merges deep layout bounding boxes with native PDFKit data, generating accurate EPUB architectures orders of magnitude faster than Python-based alternatives.

⚖️ License & Acknowledgements

• Code License: AGPL-3.0
• YOLO Framework: Ultralytics YOLOv8 (AGPL-3.0)
• Model Weights: hantian/yolo-doclaynet

⚠️ Note: The "Make it Flow" brand name, App Store presence, and UI/UX designs are the exclusive property of the author. Repackaging and publishing the app to the App Store without authorization is strictly prohibited.