Experiment with taking a red square as the input image in Stable Diffusion (running locally)>img2img.In A.I you can define positive and negative prompts as a way of being more descriptive with generative content.So starting with a square image I prompt the positive to be a green circle and the negative to be a red square, thus throwing the reference image useless.As a result I am able to get a new shape half circle/ half square infinitely trapped in its own prompted prison.