This sort of issue holds with all kinds of prompts, on both platforms. I most recently (to test Bard's image generation capabilities) was asking Bard/Gemini to

This sort of issue holds with all kinds of prompts, on both platforms. I most recently (to test Bard's image generation capabilities) was asking Bard/Gemini to generate home designs using highly specific prompts -- layout of the house, materials for the facade, window placement and style, etc -- and it was shocking how frequently it would just ignore critical pieces of the prompt, and then continue to ignore when corrected.