Haha, so, I got a preference double-choice for this, so... we get to see extra samples of how and if this worked... which is actually important to test, because most things that work 'work sometimes', not always. With GPT-3.5, I usually check stuff 10x, to confirm how often the result is consistent, and what the range is. We don't get as many 4 or image messages, so I don't check as carefully, but I 'should' if I wanted to be sure.
This one was a fail originally, and one of the two still fails.
The success prompt:
Imagine a surreal landscape with a giant, twisted tree in the center, its branches stretching towards a sky painted in vibrant shades of purple and orange. Beneath the tree, a small pond reflects the sky's colors, surrounded by unusual, colorful flowers that seem to glow softly. In the foreground, a path made of luminescent stones winds its way towards the pond. In the lower left corner, a uniquely shaped rock, perfect for perching, sits subtly. A small, puffed-up specialty-breed pigeon, with an extravagant plumage in shades of blue and green, rests comfortably on this rock, almost blending into the scene.
The fail prompt:
Imagine a surreal landscape with a giant, twisted tree in the center, its branches stretching towards a sky painted in vibrant shades of purple and orange. Beneath the tree, a small pond reflects the sky's colors, surrounded by unusual, colorful flowers that seem to glow softly. In the foreground, a path made of luminescent stones winds its way towards the pond. In the lower left corner, a small, flat rock sits ideally positioned for a perch. On this rock, a small, puffed-up specialty-breed pigeon, with an extravagant plumage in shades of blue and green, stands regally, almost blending into the scene.
The only difference I catch is the rock that worked was described more vividly, and was unusual. A 'small flat rock' maybe didn't seem worth enough attention to place right? I'm very much guessing, but I'll track that in case.