Started from a generative piece made from noise sent through video feedback, then recolored to guide AI for elements to be generated on top, and finally sent to Clip+VQGAN as starting image.