Updated 9/18/25 with Gemini Nano Banana example
When you have only a sketchy idea of the image you want, you can give a rough prompt to an image generator like Midjourney and it will please and surprise you with a detailed image. But when you do have a precise idea of the composition you want, MOST of the image generators will frustrate you with their thick-headed inability to create what you tell them to create.
Today I had a clear idea of the image wanted. I knew the characters I wanted and how they should be arranged in a composition. I would like an image generator to flesh out my composition with details and features. Let’s see how that works out. Here’s a version of my prompt.
On a modern farm, in a corral outside a barn, on the left is a very stubborn brown calf, its legs braced against the pull of a rope that is looped around its neck. On the right, a young man in worn faded jeans and a white tee shirt is pulling on that very taut rope, leaning back and straining to bring the calf across the corral. Sitting on the top of the corral fence at the rear center, a young woman in jeans and a yellow tee shirt laughs and waves encouragingly. Landscape orientation, 16:9.
Pretty clear, eh? Let’s look at some of Midjourney’s efforts.
Girl is not on the fence, calf is not braced, the rope — how many ropes are there? and none are taut, and in fact the rope is not around the calf’s neck. On one hand, it is remarkable that an algorithm, given a few words, would produce realistic people and a calf in a realistic farm corral. On the other hand, the randomness and arbitrary changes of the details demonstrate the basic flaw of present AIs: they don’t have any concept of the world (or concepts at all). They have no idea of how a rope around an animal’s neck would behave. They have no concept of “neck” or “around”.
Here’s another try by Midjourney.
Much closer. Girl still not “sitting on the corral fence” — what part of “sitting”, “on”, “fence” did it not get? — but at least the rope is taut. Or is it? The AI did manage to draw a taut rope but it is not attached to the calf. It is “looped” — around the young man. Hey, a loop is a loop, right?
Out of 16 Midjourney images, that one was the closest to the requested composition. Here’s a typical screw-up from the same group of four (Midjourney makes four images per try).
Nice lighting, ok? Girl is not sitting on the fence (but someone is sitting beyond it). She’s holding the rope, not waving. The rope looks taut. But if you look close, it is not looped around the calf’s neck. It kind of slants toward the neck and up again.
In my role as art director, I will fire this artist. I gave a quick trial at Leonardo.ai, and while it did manage to position the woman on the fence (although not in the center as requested) it failed just as dismally with the rope. Oh! And just how many legs does that gal have? Fire Leonardo and try a better system.
On to Google’s Gemini, which came much closer.
I have to admit, that is darn close to what I imagined. Gemini actually has an idea about how rope interacts with a cow’s neck and head. If the woman was looking at the action instead of at the ground, I’d have to say it is good. A solid B+ effort anyway.
Well, how about ChapGPT?
I have to give this a B. It has all the requested elements, positioned as requested. The lighting is rather drab, but I have to acknowledge a subtlety — the woman and the barn are slightly soft-focus. ChatGPT understands “bokeh”!
Two nits to pick. The rope emerging from below the neck doesn’t look right. Much worse, however: look at the woman’s right hand.
Classic AI screw-up right out of DALL-E in 2023: six fingers, no thumb! The very latest GPT still can’t do hands!
9/18/25 – Google sends an invite to try Nano Banana in Gemini. The link went to what looked like an ordinary Gemini page, but I put in the prompt from above and got back, amazingly, a pretty accurate rendition.
Definitely the best treatment of the rope: it’s in a credible halter form and looks like it’s pulling the calf. The calf, the man, and the woman are in the positions requested. You have to really dive into the nit-picking details to find something to complain about.
How many fingers in the right hand? And there’s something funny about how the rope kind of merges with his belt. But on the whole, as of now Gemini wins this little contest.
Codgerville 