TRAINING MACHINE LEARNING MODELS FOR FX ARTISTS

In this tutorial we look at a scene file that trains a machine learning model to predict the next frame of a fluid simulation. We then feed the ml model's prediction back to the ml model recursively to create the appearance of a simple smoke sim. Only the nodes of interest are discussed in the tutorial, keeping the video from getting overly long.

This video can be watched at 2X speed to just get general information on training models on volumes, or download the scene file and try it out with the video to get hands on experience. The scene file could also be a good template to start from for other volume ml models you my want to create.

Why for fx artists? The steps an fx artist goes through to do large simulations is very similar to what you will do to train a machine learning model. You prepare your data (emitters, geom, and attributes) going in, plan out the simulation configuration, and then run the simulation which for big ones could take from 1-8 hours. Finally you evaluate the output, make any needed changes, and iterate until you get what you are looking for.

that .