nuke graph next-node prediction

A Nuke plugin that leverages a locally trained machine-learning model to predict the next node based on your unique workflows.

Includes a full framework for training Graph Attention Networks (GATs) specifically for Nuke, featuring a script parser and serializers, GAT implementations, and a lightweight FastAPI server for performing model inference in a separate process.

Features

• ML Predictions: Utilizes a Graph Attention Network (GATv2) for intelligent node recommendations.
• Graph Serialization: Converts .nk scripts into structured and flattened graph data for model training and inference.
• End-to-End Training & Inference: Supports model training and predictions via a custom panel within Nuke.

Getting Started

1. Copy the contents of init.py and menu.py:

   # menu.py
   from nuke_auto_predict_node import nuke_auto_predict_node
   nuke_auto_predict_node.on_startup()

2. Generate virtual environment for inference:
   • When launching the plugin for the first time, bin/create-venv.sh will be executed.
   • The dependencies for the inference server will be automatically started.
3. Open the Prediction Panel:
   • Navigate to Windows > Custom > NodeRecommendation in Nuke.
4. Train a local model:
   • Under the "Training" tab, click Browse and select a folder containing your Nuke scripts.
   • Click Train to begin training the model on your scripts.
5. Get Predictions:
   • Once the model is finished training, click on the Prediction tab.
   • Press Ctrl + Shift + T when selecting a node to perform a next-node prediction.
     • Or enable Predict on Selection to perform predictions automatically upon selecting a node.
   • The predicted Node Type and confidence score will be displayed.
   • Double-click a node-type entry to automatically create it in the node graph.

Logs & Debugging

Logs are stored in:

checkpoints/logs/auto-predict.log

Inference logs are stored in:

checkpoints/logs/server.log

Contributing

Contributions are welcome! Please fork the repository and submit a pull request for any improvements or bug fixes.