This raises an interesting point. Less is more, and more is counterproductive. The flower, though perhaps the most creative AI could manage, is excellent. I've found that excessive instructions, even if innocuous, are prone to offend the input and output monitors, while simple carefully-worded instructions produce desired and often surprising results. In between the monitors, AI seems to take hints well.
Here's the composite results of another attempt. It makes me wonder if (1) a set of pictures might be a good prompt and (2) the misinterpretations might be more effective than the intended results. The descriptive text was "A blond teenage boy with a crew-cut wearing blue jeans and a white muscle shirt filling his automobile gas tank from an old-fashioned pump at a 1930's gas station in a rural town". AI did fairly well with the setting and the physical description (other than the crew-cut part), but it doesn't seem to understand the process. The poor kid either couldn't find the hole, found the wrong hole, or when he found the right one, had a bit of a problem getting it in. A picture is worth a thousand words, and I'm sure one could write a thousand-word story off this one. 😈