Machine learning inside Nuke has moved well beyond experimentation. With The Foundry’s CopyCat node, compositors can now train custom convolutional neural networks directly inside the node graph, solving real production problems using shot-specific data. That means ML isn’t something happening in a separate research department. It’s part of your compositing toolkit. In our new course, CopyCat for Nuke Machine Learning, Doug Hogan walks through building and refining a complete ML-assisted roto and beauty workflow from scratch, using practical production scenarios rather than abstract demos.

What Makes CopyCat Different?

CopyCat allows artists to train models on their own corrected frames and reference examples. Instead of relying on generic AI models, you can create shot-specific solutions for:

• Roto assistance
• Beauty cleanup
• Reflection and glint removal
• Targeted image reconstruction
• Patch-based fixes and consistency work

Because training uses licensed base models and your own curated data, the results are predictable, commercially safe, and fully under your creative control. The power of CopyCat isn’t just in training a model once; it’s in understanding how to iterate, refine, and deploy that model responsibly in production.

Inside the Course

This is an intermediate, project-based course designed for artists who are already comfortable with Nuke. Across eight classes, you’ll:

• Build a training dataset from a real hero shot
• Create clean reference frames for supervised training
• Run and monitor model training inside Nuke
• Evaluate validation loss and checkpoint performance
• Diagnose overfitting and stalled training
• Refine and retrain models through targeted iteration
• Extend your model to a second shot
• Package models for studio deployment

A core focus of the course is decision-making. When should you fine-tune versus retrain? How do crop size and batch size affect stability? When is a model production-ready, and when is it drifting? Rather than treating machine learning as a black box, Doug breaks down how CopyCat works under the hood and shows how to make informed, repeatable choices that scale in real studio environments.

Production-Safe Machine Learning

As ML tools continue to evolve, responsible implementation matters. This course emphasizes:

• Using licensed base models
• Training only on shot-specific data you provide
• Structuring datasets for predictable results
• Maintaining creative ownership and commercial safety

The goal is not experimentation for its own sake; it’s building workflows that hold up under production pressure.

About Doug Hogan

Doug Hogan is a Creative Technologist, VFX Supervisor, and veteran Nuke compositor with over 18 years of experience in feature films, animation, commercials, and immersive media. He spent much of his career at Reel FX, where he built and led the Compositing and Matte Painting departments on films including Scoob! and Rumble. Today, Doug works at Groove Jones, pushing the intersection of AI, XR, and traditional post-production. Doug is also the creator of the open-source Nuke MCP project, an active contributor on Nukepedia and GitHub, and host of the VFXTalk community. His teaching focuses on clarity, repeatability, and real-world production integration.

If you’re ready to integrate ML into your Nuke workflow with confidence, CopyCat for Nuke Machine Learning is available now on fxphd.