We can initiate AI with text prompts or by clicking a camera shutter. Why is the output of only one of them eligible for copyright?