How Many Midjourney Prompts Does it Take to Get the Perfect Stock Photo?

6 min readJun 8, 2023

In which the author spends way too much time in Midjourney to get to a useful image

Stock photography is generally boring, but it can be useful to get across an idea.

The promise of being able to come up with your own ideas for a stock photo — one that isn’t so boring — seems like a dream come true for those of us who regularly write for a business audience.

Case in point: I had an article where I wanted to make a comparison between poorly made patio furniture and incomplete customer feedback. The patio furniture is a potential hazard for your body, and the feedback is a hazard for business decision making.

Round 1: Perfect Furniture

I took to Midjourney with my first prompt:

“IKEA patio set that is missing several screws and bolts. The furniture looks like it is ready to fall apart if anyone sits in it. Realistic style — ar 3:2”

(As a quick aside, the last part “ — ar 3:2” means “aspect ratio of 3 to 2.” I didn’t know that Midjourney could create anything other than square images until just recently.)

Here’s what I got:

Um… I’d like to buy that patio furniture. I don’t see how it gets across the idea of poor construction.

I slightly tweaked the prompt:

“patio furniture, including table and chairs that are ready to fall apart. The furniture is new but poorly assembled. Realistic style. — ar 3:2”

patio furniture, including table and chairs that are still in immaculate condition, despite my prompt

Nope. Still immaculate.

Come on, Midjourney. I’m going for a metaphor here, and this isn’t working.

Round 2: Trying to Portray a Struggle

My next thought was that I could maybe include a teenager along with the furniture. His expression would show that this project was going poorly, so it would be more obvious that the furniture wasn’t turning out as expected.

Next prompt:

“teenaged boy sits on a patio assembling patio furniture but is clearly struggling. His face shows his confusion. Photorealistic style — ar 4:3”

Midjourney has come a long way with correcting issues from earlier versions. We’re now on version 5.1, and things are vastly improved. Here’s an image currently going around social media to show off how much better things are…

Yoda takes a selfie, as seen in each version of Midjourney. Remarkable progress in just over a year — Image by Ruben Hassid — h/t to Heather McKay for sharing

That’s pretty amazing, right?

Ok, for all the progress, there’s still some oddities in the image I asked for…

Boy looking at a tool in his hands seems to have 3 hands — What’s going on with his hand?

And this

Boy sitting at a table seems to have two legs at his torso, but the legs merge into a single leg — Does he have just one massive leg?

I tried getting some variations on the first image from the above set, and they were okay. But still not exactly what I was looking for.

And, these image aren’t better than a regular stock photo:

Also, this teenager seems more discouraged than confused. I want to give the poor kid a hug and a pep talk.

A few more tries, but this time I removed any description of the boy’s expression:

“teenaged boy sits on a patio assembling patio furniture but is clearly struggling. Photorealistic style — ar 4:3”

Honestly, the above set isn’t too bad. They just don’t convey the sense of humor I was trying for.

In each image, the boy is building, and there is no sense of incompetence or struggle.

I went for an older teenager:

“older teenaged boy struggles to assemble patio furniture. Photorealistic style — ar 4:3”

Again, not terrible, aside from the weird arm thing happening in the top right image. But not what I was looking for.

Round 3: Things Just Get Messy

Maybe I wasn’t getting across the hectic sense of it all.

“teenaged boy struggles to assemble patio furniture. pieces of furniture are scattered on the ground. Photorealistic style — ar 4:3”

teenager sits in the middle of a mess while trying to build patio furniture

This was getting closer. Things are obviously not going well for this teenager.

I focused in on the first image and tried variations.

boy works on patio furniture, but Midjourney has placed him in an impossible squat where he seems to be sitting but no chair is beneath him — This is promising…

What about the fourth image?

Maybe not apparent from a zoomed out view, but the facial features are odd on these images

So I tried a higher quality version of the first image… only to figure out that this boy has mastered some sort of squat that requires no chair:

boy squats in mid air while trying to build furniture — Hard to unsee once you notice

Round 4: Get Rid of the Teenager

You may be tired of seeing my attempts at this point.

I definitely gave up a few times in the process of trying to get my prompts correct.

Still, I gave it a couple of more attempts before finally finding a winner.

“broken patio furniture”

(Not creative, I know. But maybe I was overthinking it.)

These images were interesting, but this was more about decay than it was improper assembly.

Finally,

“patio furniture in pieces, not yet assembled”

What is happening with some of this furniture? That is some bizarre stuff! But image #1…

I had a winner.

patio furniture in pieces, not fully assembled

Some Thoughts

I know some artists are really worried about the advent of tools like Midjourney. I was able to spit out a ridiculous number of images in a relatively short amount of time, and no human would have been capable of producing that much content.

But…

There was so much trial and error to get at what I wanted. Even in the final image, I got to a “close enough” state that I stopped.

I get the opportunity to work with human designers everyday, and the art-requesting process requires maybe two rounds to get to what we’re looking for.

Artists and designers can still do many things that these systems cannot.

Be proud of that. Lean into that. Convey emotion in what you do.

It’s Easy, Right?

Many advocates of AI systems promise that these tools will solve all your problems and make your life easier, but there’s definitely a learning curve here.

In the same way we had to learn to use Google (phrase your search query just like this for the best result), we have to learn to speak the language of these systems.

It’s a necessary skillset, so I definitely recommend taking the time to do it. Just be ready for some trial and error.

Maybe We Can Make Business Articles More Interesting

If we have enough time for prompts, maybe we can add more of our own emotions into these elements.

I think we could all do with some more entertaining education.