A trick used by many "prompt helpers" for AI-generated images is spam out a load of tiny and unnecessary details, half of which will get ignored in any given generation because they tend to conflict with each other. This seems to be consistent with your own experiences at suggesting a single object, but it might be worth checking if the mechanism by which this causes creativity is the same in both cases, and if so is there a point on the continuum between "one detail" and "absurdly many details" that produces the most creative results.