In what ways has image generation been solved? Prompt blocking is about the only real effort I can think of, which will mean nothing once open source models reach the same fidelity.