Prompt fidelity of ChatGPT4o / Dall-E3 text-to-image visualisations