Addressable LEDs in some ornament. A neural network drives the signals. It defines timings and selects what LEDs should be on or off. A camera feed goes to another network which provides feedback for the first network. I propose, instead of designing special algorithm for switching lights or crafting special reward function for reinforcement learning system, it will make decisions based on the visuals it perceives from the camera.

It essentially creates a closed-loop system where the visual input guides the lighting decisions.