Too many pieces of software involved... original is realtime generative shapes and text made in VVVV... then record some video from that... run that video through a neuralstyle transfer (fast feedward tests), then color over correct to give it a vibe in AE. Overkill for such a basic vj clip that doesn't even loop, trying to get a workflow that allows me to do this neuralstyle stuff in realtime, and am getting pretty close (although it doesn't look as cool as the stuff with forward and backward optical flow pre-calculated).