GAN Based Virtual Production
This untitled piece was an experiment in combining generative AI and virtual production with a classic animation aesthetic, i.e. Disney’s Multi-Plane Camera.
Assets are generated by various trained StyleGan models using a custom built StyleGan designed to integrate with Unity.
During scene creation, the artist can generate new 2D assets on the fly, or create morphing animations to create a whimsical watercolor animation.
The animation and datasets where all created in unity with a watercolor shader being used as the final post processing effect.
Unity is used to generate a dataset large enough to train a StyleGan, roughly 10k images.
The trained style gan is then capable of generating novel assets and animations.
A bespoke Unity interface is created to create shots, post effects, as well as interact with trained StyleGan Models via a web API.