Logik Live #57: Make your own ML Tools with Nuke's Copycat

I’ve had a blast experimenting with Nuke’s Copycat, which lets you train your own Machine Learning models and use them as compositing tools. I’ll show you Copycat from a Flame Artist’s perspective: How it works, different use cases and how to setup Pybox so you can use your trained Copycat models from right inside of Flame!

Sunday July 25th @ 2pm ET on YouTube. For a chance to win one of our weekly prizes register at http://www.logiklive.com

After the show, I’ll be doing an exclusive session for our Patreon Patrons where I’ll show more Copycat examples. If you’d like to check it out please consider becoming a Patron for as little as $5/month by signing up at http://www.patreon.com/logiktv

Logik Live is proudly sponsored by CineSys.io and AJA Video Systems


Hi Andy

This sounds really interesting. Would it be possible to do the whole operation in flame rather than hop into nuke?


Hi John,

I think the short answer is no. It might be possible to do some type of advanced python scripting to get your training images into a nuke script, kick off the training and make your Pybox script, but that’s beyond my level of experience. When you see the presentation I think you’ll see that it requires the bare minimum of nuke skills (adding nodes, connecting them, reading/writing images). Plus, there’s value in testing your trained models in Nuke before setting it up for Pybox.



Sorry. What I mean can you do the training in flame? Didn’t @tpo do that?

Anyhoo, I am looking forward to the broadcast. I was trying to get a model for the cartoon face working in flame the other day but couldn’t make head nor tail.

Gotcha. I think what he was doing was using Pybox to generate mattes off of a trained model. When he was on Logik Live in January he was using a pre-released version of Copycat and had to keep that part of the process hidden. That being said, I haven’t dived any deeper into Pybox beyond the basic input-render-output, so it may be possible.