most neural networks would draw even worse pictures given the same prompts