However, what stable diffusion does bears no resemblance to "add 1 to each RGB value" and bears substantial resemblance to the human process of "learning". Wit

However, what stable diffusion does bears no resemblance to "add 1 to each RGB value" and bears substantial resemblance to the human process of "learning". Without litigating if the model is "learning" in a human analogous way, instead I think the burden is on showing that the model is "substantially copying" and to an equal or greater degree than what would be needed to consider a work derivative.