MageML

🛠️ Version 1.0.0 BETA — Initial Public Release

Check out the latest changes

---

📚 Documentation

• 📝 Changelog
• 📦 Installation Instructions
  • 🛠️ Building from Source
• 🤝 Contributing Guide
• ❓ FAQ

---

🚀 Features

MageML is a blazing-fast, node-based graph editor for video and ML pipelines.

• Visual, drag-and-drop node editor for building AI/video flows in seconds
• ONNX model support – run any compatible ML model, no code required
• Chain arbitrary nodes (decode, filter, ML, encode, and more)
• High-performance, pure C++/CUDA backend for real-time throughput
• Export for full automation workflows

For More Information on available nodes, please see Nodes

Powered by CeLux (fast video/tensor core), ONNX Runtime, PyTorch, and imgui-node-editor.

---

🎥 Watch Demo Video

---

---

⚡ Quick Start Guide

MageML is a real-time, zero-copy machine learning graph editor for video and tensor pipelines. Here’s how to get up and running fast:

---

🧩 1. Download the Latest Release

Grab the latest release from the Releases page.

---

🛠️ 2. Extract Files

Extract the Zip File to your preferred location.

---

🚀 3. Launch MageML

Start the app from a shortcut or directly via MageMLGUI.exe within the root folder you extracted into.

---

🧠 4. (Optional) Add Models

Place any .onnx models into the models/ folder. Some starter models may already be included.

---

🖱️ 5. Start a Project

Use the UI to load a template graph or create a new one. Drag and connect nodes to begin building your pipeline.

---

🧭 UI Overview & Tips

• Top Left Panel:
  Controls for Live and Basic modes:

  • Live Mode: All nodes recalculate automatically on changes. Video pipeline will not push unless running.
  • Basic Mode: Only Video Reader and Preview nodes compute. Ideal for layout/design without processing overhead.

  You can toggle modes without running the graph to safely test behaviors.

• Run / Stop:
  Starts/stops the video processing pipeline. You can also modify node parameters during runtime for real-time feedback.

• ⚠️ Encoding Tip:
  Avoid scrubbing during encoding. Let the graph play to completion.
  Be sure to click Finalize in the encoder node to ensure the file is written!

---

🧩 Node System

• Video Reader Node:
  Enables timeline control, Live/Basic toggle, and playback.

• Active Node List:
  Located below the main toolbar. Double-click to rename nodes.

• Node Inspector Panel:
  Appears below the node list. Displays the selected node’s parameters and controls.

• Right Click Actions:

  • On nodes, see input/output tensor values (great for debugging)
  • On graph background, spawn new nodes by category
  • On pins/links, access connection options

---

💡 Performance Tips

• For max speed: Remove Preview Nodes when not needed.
• MageML is zero-copy: Tensors only clone during Preview or Encoding. All other ops run in-place.

---

⌨️ Hotkeys

Shortcut	Action
Ctrl+O	Open graph
Ctrl+S	Save graph
Ctrl+Shift+S	Save as
Ctrl+C/V/X	Copy / Paste / Cut node
Ctrl+A	Select all
Ctrl+Shift+A	Deselect all

Graphs are saved as .json, and reusable templates can be added to the templates/ folder for quick access via the File menu.

---

🎨 Style Menu

Use the Style menu to customize the look of:

• Node shapes, colors, and font size
• Link styles (flow map, wire thickness)
• Grid layout, zoom, snapping

---

📄 License

This project is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0).
See the LICENSE file for details.

---

🙏 Acknowledgments

• CeLux – Fast video/tensor core
• PyTorch – For tensor ops and CUDA support
• ONNX Runtime – Universal ML inference
• imgui-node-editor – Graph UI
• vcpkg – C++ dependency mgmt

---

For questions, ideas, or just to show off your graph, join our Discord!