10 of My Most Popular AI Image Series (+Prompts)

Revisiting my Midjourney and DALL-E 3 experiments along with the prompts I used.

Happy Thursday, art artificers,

In case you didn’t know, on top of writing a lot about Midjourney, I also actively use it and other text-to-image models to experiment with random visual ideas that pop into my head.

That’s how I ended up above the threshold of 15K+ image generations that gave me access to Midjourney’s Alpha website already last December.¹

“Daniel, we need to talk about your text-to-image problem”

I often share the resulting image series on Reddit or Facebook. Sometimes, enough people enjoy them that they get a lot of shares and engagement.²

When that happens, I tend to get questions about the prompts I used to generate the images. People assume that these prompts must be quite advanced, using lots of modifiers and other voodoo sprinkled in.

I want to dispel that myth once and for all.

So today, I’ll share my 10 most popular AI image series and the exact prompts I used to get them.

You’ll see that there’s very little dark magic involved.

A few thoughts on AI image prompting

Just a quick aside.

I’m not a fan of the term “AI artist.”

I never thought of myself as one.

If you’re not using AI images as a starting point in a longer workflow and simply stop at requesting an image from Midjourney, “artist” is way too strong a word.³

Source: Know Your Meme

A better descriptor for the role most of us take with AI images is that of a prompter-curator. You prompt the model to generate a bunch of images, then you curate the results to handpick the best ones.

That’s typically it.

And when it comes to the first part of that role—prompting—I advocate for a less-is-more approach. I’ve been against “splatterprompting” since January 2023.

More recently, I argued that text-to-image models are now so good at natural language that you can just get help from LLMs to prompt them.

My ever-expanding list of the best Midjourney prompts follows this philosophy. I only share short prompt modifiers but never long-winded prompts full of questionable stuff like “award-winning,” “ultra-realistic,” etc.

Here’s the process I recommend when working with text-to-image models:

1. Start with a simple, minimum viable prompt and only expand it if necessary.

2. Be ready to ask for many iterations before you get the image you want. Many of my series required dozens of rerolls for each image.

Most importantly, just have fun with it!

Without further ado, here are 10 of my popular image series, grouped by their Midjourney “era”: V4, V5, or V6. (Plus one DALL-E 3 collection.)

The post might get cut off in some email clients. Click here to read it online.

Midjourney V4

Funnily enough, despite V6 being a massive upgrade, some of my favorite series come from the V4 era.

Maybe that’s because text-to-image AI was still new and exciting and I generated a lot more images at that time.

Here they are in alphabetical order.

1. Cats in Hats: Worldwide Tour

This started as a random thought: “What if The Cat In a Hat, but the cat is a country and the hat is country-appropriate headwear?”

But then it ended up being a bit personal.

I was born and raised in Kharkiv, Ukraine, only moving to Denmark at the age of 14. My mom and older brother still live in Kharkiv. Some of my childhood friends are defending the country on the front lines.

(I’ve been writing about Ukraine elsewhere, starting with Russia’s annexation of Crimea in 2014 up to Russia’s full-scale invasion in 2022.)

So the Ukrainian cat wearing a military helmet wasn’t just a gimmick to me.⁴

Sample images (click an image to see it in full resolution):

Prompt structure:

portrait of [country] as a cat wearing a [type of headwear]

For Ukraine, the exact prompt was:

portrait of Ukraine as a tough cat wearing a modern military helmet and goggles

Notes:

Midjourney really struggled with some of the headwear, so for many countries, I didn’t get quite as close as I wanted. Also, the “United Kingdom” one should have been “England,” especially since I’ve had Scotland as a separate cat.

2. Cyberpunk Farm

This was a similar collection of animal faces wearing cyberpunk goggles. It uses one of my all-time favorite Midjourney modifiers: “blacklight”

Sample images (click an image to see it in full resolution):

Prompt structure:

cyberpunk [animal] portrait, goggles, blacklight

So the rabbit one is simply:

cyberpunk rabbit portrait, goggles, blacklight

Notes:

As you’re hopefully starting to notice, prompting really isn’t rocket science. Even during V4 days, simple and specific prompts were often enough.

3. Seven Deadly Beans

The idea was silly wordplay: Let’s have Mr. Bean act out the seven deadly sins and call them the “Seven Deadly Beans.”

The fun came from figuring out which situation would best represent the sin in question while also being possible for Midjourney to render.

Sample images (click to see in proper aspect ratio):

Prompt structure:

Mr. Bean [scene description], medieval Biblical illustration

For Sloth, the prompt was:

Mr. Bean lying and sleeping on the floor surrounded by trash, medieval Biblical illustration --ar 3:2

Notes:

Yes. That’s Comic Sans. No, I have no regrets. Read “I’m Comic Sans, Asshole,” to hear from the font itself.

Each image in this series took many rolls to get right. I have hundreds and hundreds of discarded Mr. Beans in my Midjourney archive.

4. The Real Spaghetti Western

“What if spaghetti Westerns were literally made out of spaghetti?” I asked. Then I forced the poor Midjourney to show me the answer.

I really like this collection because it has a narrative: A lone outlaw arrives in a small Western town to settle an old score. He finds his foe in a saloon and a fight breaks out. In the heat of the moment, our outlaw ends up shooting the sheriff and escaping into the sunset, unsuccessfully chased by the deputies.

Sample images (click to see in proper aspect ratio):

Prompt structure:

[scene description], made out of spaghetti

OR

spaghetti sculpture of a [scene description]

So for the image of the cowboy on a horse:

spaghetti sculpture of a cowboy riding into the sunset, view from the back --ar 3:2

5. 16-bit Sunsets

I always loved the way Midjourney V4 rendered 16-bit pixel art. So I asked for a bunch of sunsets over different locations and countries in that style.

The prompts were very simple, as you’re about to see.

Sample images (click to see in proper aspect ratio):

Prompt structure:

sunset [scene / location], 16-bit

The first image above was simply:

sunset in Vietnam, 16-bit --ar 3:2

Notes:

To this day, Midjourney V4 is the best at recreating the 16-bit vibe. So if you’re after similar images, I recommend using V4 by either switching to it in the menu or appending --v 4 to your prompt.

Midjourney V5

One of the highlights of Midjourney V5 was its ability to generate believable photographic images. So my series from this era often revolved around photos.

6. Awkward Superhero Yearbook Photos

Have you ever wondered what a young Hulk would look like in his awkward yearbook photo? You never have? Is it just me?

Ah well, here we go:

Sample images (click to see in proper aspect ratio):

Prompt structure:

young [character], [additional details], worst/awkward yearbook photo

So Wonder Woman’s bad hair day was just that:

young Wonder Woman, bad hair day, worst yearbook photo --ar 4:5

Notes:

Yes, I accidentally wrote “Spiderman” instead of “Superman” when labeling the images on Reddit.

7. Victorian Families and Their Strange Pets

Midjourney V5 also made rather believable vintage photos.

So I immediately wanted to see if I could place sci-fi robots and modern characters like Yoda and Groot into old photos with Victorian families.

I could.

Sample images (click to see in proper aspect ratio):

Prompt structure:

vintage photo of a Victorian family with [character]

For the middle image, the prompt was:

vintage photo of a Victorian family with their Mothra

Midjourney V6

With the arrival of V6, Midjourney was suddenly able to follow much longer prompts written in natural language. So we could start using more detailed scene descriptions.

8. Harry Potter in Soviet Russia

The popularity of this series took me by surprise. It hit 2,5 million views and 15K+ upvotes on Reddit and tens of thousands of engagements across different Facebook groups.

I just wanted to see what Harry Potter would look like if it was set in Soviet dystopia.

What if Voldemort was a tracksuit-wearing gopnik? What if Dolores Umbridge was a mass surveillance officer?

For answers to these questions, look no further.

Sample images (click to see in proper aspect ratio):

Prompt structure:

Scene from Harry Potter filmed in Soviet Russia, [scene or character description], cinematic movie still

Voldemort was the result of this prompt:

Scene from Harry Potter filmed in Soviet Russia, Lord Voldemort wearing a tracksuit and looking into the camera, dystopian setting, cinematic movie still

9. Rock Star Wars

For this series, I wanted to use Midjourney’s newfound ability to render short text.

So I gave Star Wars characters goofy artist names and merged “Rock Star” with “Star Wars” to create vintage posters advertising their performances.

Sample images (click to see in proper aspect ratio):

Prompt structure:

[scene description]. Vintage poster with the text "[Text]"

For Leia, the prompt was:

Princess Leia playing a synthesizer. Vintage poster with the text "Princess Slaya" --ar 4:5

DALL-E 3

I didn’t create nearly as many series with DALL-E 3. (Unless you count the regular AI Jest Daily cartoons.) But I did test its ability to spell when it first came out.

10. Misunderstood Superheroes

For this collection, I imagined a hapless graphic designer tasked with creating movie posters for superhero movies but somehow managing to completely misunderstand the heroes’ names and the meaning behind them.

Sample images (click to see in proper aspect ratio):

Prompt structure:

Tall movie poster [scene description]. The title says [text].

The Black Pan Sir poster used this:

Tall, atmospheric movie poster illustration set in a dim-lit, moody ambiance. A sophisticated man stands confidently in the center, dressed in elegant attire, holding a sleek black frying pan. The contrast between his polished look and the kitchenware hints at a humorous twist. The title 'Black Pan Sir' is clearly and prominently displayed at the bottom.

Notes:

The prompts here were written by ChatGPT (GPT-4) itself, which followed the default system prompt from OpenAI. It often tends to include non-visual details that I doubt have any impact on the resulting image (e.g. “hints at a humorous twist”).

I later learned to clean those up manually, but ChatGPT took the wheel for this series.

Over to you…

I hope I managed to prove that it’s possible to turn your ideas into fun images without relying on complicated prompts.

Do you also experiment with AI images, AI video tools, or other visual generative AI? If you do, what’s your go-to text-to-image model or text-to-video site?

Feel free to share some of your collections. Send an email to whytryai@substack.com or leave a comment below.

You can also message me directly:

For more of Midjourney and DALL-E 3 experiments, visit my Behance:

1. The number is now actually 20K. I have issues.

2. I’m sure some of you have subscribed to Why Try AI? after coming across one of my online collections.

3. And that’s without getting into the debate about text-to-image models being trained on the work of traditional artists.

4. Fun fact: Ukrainian soldiers are endearingly nicknamed “kotiki”—something between a “kitten” and a “male cat.”