Most days, my spare time is often spent making handmade stuff. Recently, that changed when I got the most amazing gift for my birthday: an invitation to use Midjourney, the beta AI-powered tool for generating digital art. So far, my fave type of synthetic images to make are Midjourney portraits.
After spending a couple of weeks getting to know the ins and outs of using this rad technology, though, I’ve discovered that rendering decent-looking portraits seems to be one of the most challenging output types for this app to get right.
That’s why I wanted to write a blog post to help others refine their prompt-writing process with Midjourney so they can generate the best possible images (and avoid repeating my mistakes).
A Portrait Generating Tool for Non-Artists
For folks like me with very little natural ability to draw, the idea that I can turn what’s in my head into art without a paintbrush is exciting. That said, it’s kind of boring just to push a button and get something beautiful in one try.
The collaborative nature of using Midjourney AI is a large part of its appeal. Having more back and forth to get a generated image to look good (within a community of others from around the globe also trying to figure it out) makes the process more rewarding and fun. (Of course, if you’re self-conscious, I should mention that you can run /imagine commands by chatting directly to the Midjourney bot. But what’s the fun in that?)
Keep reading to get secrets for writing prompts that will create the most compelling Midjourney portraits, answers to common questions, and view tips for doing minor clean-up in Photoshop and other apps. Plus, see examples of Midjourney images with the inputs I used to make them.
Estimated reading time: 35 minutes
Midjourney AI Basics
What Is Midjourney AI?
Think of Midjourney as an “art-making robot” that you talk to on the web community Discord in order to generate images using artificial intelligence.
How Do You Use Midjourney AI?
The tool is still in closed beta, so its nuts and bolts are changing daily. Once you have access, using the app is pretty simple.
All you have to do is type in specific prompts to guide the program’s visualization creation process.
Can Midjourney Create Art From Any Word Combinations?
It’s essential to keep in mind that Midjourney is like other AI tools regarding how it processes language. From what I’ve observed, it won’t “read” text the same way you and I do.
While we might be keen on a story, you’ll likely get better results if you write AI prompts that are straightforward and clear. That’s because Midjourney’s text-to-image bot appears to segment anything you input.
It takes each of these individual “tokens” as separate commands and uses them to process and output different AI-imagined images for you. So when you write in sentences to describe what you want, it will try to figure out what you want and then pick and choose which words from your blurb to use.
That said, sometimes the most incredible stuff you’ll see come out of Midjourney is content that another user made by pasting in their favorite song lyric, a line from a movie, a verse of poetry, or random word combos.
For example, my niecelet’s self-selected nickname is “Nugget Nightmare.” It’s silly and doesn’t mean much when you read it, but the portrait that Midjourney created for that input was crazy cool.
If you have GPU time to spare on processing randomness, it might be worth it to see if you can generate something interesting.
What Are Some Ideas to Help you Get Started?
First, my advice would be to pick a subject for your Midjourney portrait. The more well-known the individual, the more images the AI tool will have to mine as inspiration.
For example, if you decide to try and generate a portrait of one of the ladies from the Real Housewives of Dubai, it will be much harder to produce great results since they aren’t as well known as someone like Albert Einstein or Santa Claus.
Who you pick as your muse and their fame level (aka, how many pictures of them are available online that Midjourney AI can reference) will influence the quality of your results. I also read that its image source databases, as of right now, don’t go beyond 2019.
This means, for example, that the bot doesn’t have any visual references for the craziest current events, such as the recent Amber Heard and Johnny Depp trial. If you want Midjourney AI to generate a rendering that shows a “mega pint” of wine, it’s going to be on you to feed it the right words in your input to get a fitting image.
Second, I like to think about what type of image I want. For instance, if I am going for something fantastical, I might combine my subject with a term for something that’s visually iconic.
Some Examples of Words That Help Generate Interesting Outputs
- Action Figure
- Drag Queen
Most of the time, though, I enjoy seeing what I get when I create a portrait by matching a person to an art movement, artist, or painting style.
Third, I like to figure out what to try by imagining which treatment or technique would best complement the person and fit with the look I want for my output.
Do I want a hyper-realistic graphic or a more conceptual image? If it’s someone with crazy curly hair, a rigid or geometry-based style like pixel art format isn’t likely the best fit, but an ink painting might work nicely.
Inspiration for AI Prompts
Jumpstart your AI combinations by mining this list of painting styles, art movements, formats, treatments, techniques, and time periods for words to use for inputs to create art with Midjourney:
- Abstract Expressionism
- Acrylic Painting
- Arts & Crafts
- Art Deco
- Art Nouveau
- Atelier Populaire
- Ballpoint Pen Art
- Belle Époque
- Charcoal Drawing
- Chromogenic Color Print
- Conceptual Art
- De Stijl
- Der Blaue Reiter
- Double Exposure
- École des Beaux-Arts
- Encaustic Painting
- Exquisite Corpse
- Gelatin Silver Print
- Glass Painting
- Gond Painting
- Harlem Renaissance
- Ink Drawing
- Kalamkari Painting
- Line Drawing
- Lomo Photography
- Madhubani Painting
- Neue Sachlichkeit
- Oil Painting
- Old Master
- Op Art
- Paint Splatter
- Paper Cutting
- Pencil Sketch
- Phad Painting
- Pixel Art
- Pop Art
- Trompe L’oeil
- Warli Painting
- Woodcuts / Wood Block Prints
- Works Progress Administration
To keep your inputs from getting too complex for Midjourney, I try to stick to only one or two from the list above so that it doesn’t get overwhelmed by trying to combine too many ideas into one picture.
Similarly, you can also try to see what you get if you combine your subject with a famous work of art (e.g., Mona Lisa (see example above), American Gothic, Girl with a Pearl Earring, Portrait of Adele Bloch Bauer I, etc.).
Alternatively, you could pair your subject with the name of a famous portrait artist. Below is a list of some of my favorite well-known artists for their portraits.
Well-Known Portrait Artists
- John Samuel Agar
- Sofonisba Anguissola
- Pietro Annigoni
- Giuseppe Arcimboldo
- Francis Bacon
- Elizabeth Bakewell
- H. Walter Barnett
- J. M. Barrie
- Vanessa Bell
- Giovanni Bellini
- Gustave Caillebotte
- Paul Cézanne
- Chuck Close
- Petrus Christus
- Joseph Collyer the Younger
- Agnolo di Cosimo
- Jeanne Bernard Dabos
- William Daniell
- Jacques-Louis David
- Edgar Degas
- Eugène Delacroix
- Otto Dix
- Emilio P. Fiaschi
- Lucian Freud
- Roger Fry
- Paul Gauguin
- Artemisia Gentileschi
- Shadi Ghadirian
- Vincent van Gogh
- Francisco Goya
- George Henry Harlow
- Jean Auguste Dominique Ingres
- Frida Kahlo
- Walter Keane
- Gustav Klimt
- Pyotr Konchalovsky
- Jan Lievens
- August Macke
- Édouard Manet
- Henri Matisse
- Quentin Matsys
- Anton Raphael Mengs
- Elizabeth Peyton
- Pablo Picasso
- Pierre-Auguste Renoir
- Rembrandt van Rijn
- Norman Rockwell
- John Singer Sargent
- Jenny Saville
- Egon Schiele
- Amy Sherald
- Cindy Sherman
- Ralph Steadman
- Gilbert Stuart
- Diego Velázquez
- Johannes Vermeer
- Leonardo da Vinci
- Andy Warhol
- Kehinde Wiley
- James McNeill Whistler
- Grant Wood
- Andrew Wyeth
- Craig Wylie
- Jonathan Yeo
Want to go beyond portraits? The Disco Diffusion Artist Study Database is a great Google Sheet to refer to for even more artist inspiration.
How Do You Get Access to Use Midjourney?
Right now, Midjourney AI is private, so you can’t get in to use it without an invitation. But don’t give up hope. You can apply for access at midjourney.com (that’s how I got in).
Although I can’t do so, some members can “adopt” new members and share invitations. They give you the ability to make some free images, but—if you really want to go to town—know that right now, the paid plans for using the tool start at $10/month USD.
7 Tricks for Making the Best Midjourney Portraits
There’s a saying that goes, “garbage in, garbage out,” that applies to using Midjourney to make portraits. Though there is a little bit of luck when you use the tool, what you input definitely has an effect on what you output.
And, even when you get to something close to what you want, sometimes the results are still a little “uncanny valley.”
1. Use Repetition
When you see exceptionally unique Midjourney portraits, the chances are good that the image was created after a very patient user used the tool to generate many different iterations.
The V1, V2, V3, and V4 buttons are your friend. Use them until you end up with one image that is close to the visualization you had in mind.
If that doesn’t work and your results are off, keep tweaking your input. Make subtle additions (or subtractions) one at a time to see how each change helps to vary (and hopefully improve) your outputs.
2. Copy from the Best
Working in Midjourney is a collaborative process. You can see what everyone is doing, and they can see what you are doing.
All work is shared out in the open. That means you can build on someone else’s generated art and they can push out variations from what you initialize with the AI bot.
One of my favorite sources of inspiration for making Midjourney portraits is to look at example outputs from users whose work you enjoy and see what words they are using in their prompts. Then, I copy key elements from their wording to create my own prompts that I can use to iterate different visualizations.
How to Get More Sharp-Dressed AI Portrait Subjects Clothing can sometimes be an issue for the Midjourney art-drawing bot. The good news is that with tightly-cropped portraits, the “not-being-great-yet-at-rendering-clothes” issue is less of a problem than with other types of compositions where you’d see a person’s full body. However, if goofy renderings of clothes irritate you, one workaround (to avoid odd collars, weird turtlenecks, and other shirts with strange necklines) is to tell Midjourney in your AI prompt that you’d like the subject of your portrait to be draped in robes or wearing another kind of loose clothing item. For instance, if you were to do a portrait of Bryan Cranston in character as Walter White from Breaking Bad, you might want to tell Midjourney to dress him in a baggy yellow hazmat suit (with the hood down, of course, so you can see his face and know it is in the portrait).
3. Try a Mash-Up
Unexpected combos can lead to great Midjourney portraits. There are a couple of easy ways to do this, but two of my favorite go-tos use these prompt formulas:
- person one : : person/animal/thing two (e.g., rihanna : : llama)
- half person one, half person two (e.g., half marilyn monroe, half charlie manson)
4. Put in an Emoji
Putting an emoji into your input is a great way to say a lot in just one character. It’s not easy to use words to describe what someone looks like when they are being flirty, but the wink emoji is a great way to get that across.
Emojis can also be a great way to provide background details to Midjourney about what your want the general mood to be for your portrait. A couple of my favorites that I like to add to the end of my prompts include the Milky Way and “high voltage” lightning bolt.
5. Include “Raphaelite”
Adding “Raphaelite” to any input is the quickest and easiest way to get more realistic and beautiful Midjourney portraits. This can make even general posts in the style of another artist look more effervescent and gorgeous, e.g.:
- oprah winfrey : : raphaelite : : in the style of john singer sargent
What to Do for More Detail Earlier in the Iteration Process Another great way to ramp up your results is to add “high quality” or “4K” to the end of your prompt text. This tells Midjourney’s text-to-image tool that you’d like less rough outputs from the get go for your input.
6. Take Advantage of the Upscale Button
Before you make variations, consider doing an upscale first. That extra bit of visualization effort can sometimes get you where you want quicker than just making additional random variations.
If it’s still missing the mark, you can make variations off of your upscaled image to get even better new iterations.
Benefits of Light Upscale
Sometimes, you don’t want all of the additional detail that a normal upscale will provide. In those cases, light upscale can be a helpful option for upgrading your output without compromising the integrity of the variation you are looking to take elevate.
7. Think About Time Period
Do you want your Midjourney portrait output to look futuristic? From the 1980s? From the 1680s? Adding time period information to your prompt is a great way to help the tool know details about how to make decisions about your portrait subject’s hairstyle, clothing, and more.
Another Way to Add a Sense of Time A variation on the idea of time is to add “baby,” “toddler,” “young,” “teenage,” “middle-aged,” or “old” as a descriptor for your prompt that you use to generate an AI portrait. For instance, we all know what Justin Bieber looks like doe-eyed and babyfaced as a preteen and young adult, but what will his appearance be like at age 80? If you want to know, write an input that includes “old” or a similar adjective (like “elderly” or “senior”) as a descriptor. This will tell the Midjourney bot that you want to generate an image of what the Biebs will look like when he’s getting up there in years.
8. Add in “School Photo” or “Mugshot”
It’s more powerful to get specific with your prompt instructions. “Portrait” works as a default to tell the text-to-image AI bot that you want to generate a head-and-shoulders image, but more descriptive phrases can give you great results, too. Some of my favorites to add in include:
- school photo
- yearbook photo
- driver’s license photo
These simple words in your input let the clever AI tool know that you want a picture of your subject looking straight at the viewer.
The acronym “KISS” (or keep it simple, stupid) is a good rule of thumb for writing AI prompts. A common newbie problem when using Midjourney is to overstuff their input text. If you give any AI tool too many conflicting ideas to consider during generation, it gets overloaded.
To avoid causing confusion during the process of generating a new visualization, stick with terse, clear descriptions of what you want, e.g.:
- dolly parton : : portrait : : close up : : paper doll : : circa 1982
Brevity and directness are good. However, it also helps to be specific.
The official quick start guide includes more helpful tips.
For instance, its authors recommend avoiding the complexities that result from using negatives (e.g., something without something else). Instead, they recommend sticking with positive commands in your AI prompts for generating art.
If you do want to try excluding something from your rendering, see the FAQ below for tips on using the “–no” command in your prompt text.
Preventing Off-Putting Off-Balanced Outputs Midjourney’s AI sometimes will overcomplicate things even if your prompt is written in an uncomplicated way. I’ve output human faces with extra mouths, unnatural noses, lopsided eyes, etc. Though generating multiple generations can help you finally land on a piece of art that looks good, one way to help Midjourney along is to add the word “symmetry” or “symmetrical” to your AI prompt text.
Bonus Trick for Creating Midjourney Portraits
I’ve always thought of my Nana as a very elegant woman. She always had her hair and makeup done perfectly, even though with her natural beauty she needed neither. I took a trip to Prague before she died. On that vacation, I visited the Alphonse Mucha Museum.
My Nana loved seeing my pictures of his work, so I wanted to use Midjourney to create a Mucha-inspired portrait of my Nana. Here’s a step-by-step rundown of what I did.
How to Use a Source Photo with Midjourney
I first picked the perfect image of my Nana, one where her face was isolated so that the tool could easily make out her features. After that, I uploaded the photo to my website to have a URL to include in my prompt. To get the look I was going for, I kept making variations until the program generated an image with the right “bones.” Also, you can use a suffix with the URL for the image you upload (e.g., adding –iw 1.8 … this additional command speaks to the “image weight” of your supplied reference graphic so that Midjourney can use it as a target).
This tiny addition tells the bot to use your provided picture as a significant influence for Midjourney’s derivative portrait output. You should place it at the very end of your prompt and include your source image URL at the start with additional tidbits in the middle).
In short, taking the time to add this extra direction to Midjourney’s AI tool is an excellent way to get results closer to features in your initial photo. For copyright reasons, it will never copy a supplied image exactly.
I made several different variations of my favorite from the bunch, then upscaled it, ran upscale max, and finished with upscale light. These steps helped refine the image further and remove some of the odd artifacts around the eyes and mouth.
I finished with some Photoshopping to clean up the rough edges and fix her nose.
Stop (Some) AI Art Messes Before They Happen
Sometimes, the initial upscale feature in Midjourney helps you get what you want.
But, other times, it does too much, providing extra detail that detracts from your generated image.
Here’s what to do to limit the potential for AI art messes.
Add “–stop 80” (as a parameter at the end of your initial prompt).
The addition of this short command in your prompt can act as a pretty decent solution.
After you get a variation you like, upscale it as usual.
Then, do a light upscale followed by a light upscale max.
Including “–stop 80” from the get go often is just enough to help prevent the text-to-image tool from going too far.
I’ve found this tiny addition (as well as “symmetrical”) in your first prompt input helps eliminate some unnecessary or distracting details, like bizarre nostrils, etc.
Tips for Enhancing Midjourney Portraits
I stick with Photoshop for enhancing my Midjourney portraits. This process is often referred to as “overpainting.”
It requires that you have some skill with using the program, not to mention an Adobe Creative Cloud license, to use this program to tweak faces, fix eyes, repair wonky ears, and remove strange waves of hair.
If you are looking for alternatives, you can sharpen details and fix small problems with easy-to-use and affordable deep fake apps, such as:
curating a Collection of Midjourney Outputs
Do androids dream of electric sheep? If they have access to Midjourney AI, those bots mentioned in the title of Philip K. Dick’s famous novel wouldn’t have to fall asleep to picture wooly beasts made of circuit boards and gears.
This fantastic tool allows anyone with a computer and web connection to become an AI-dreamer. Here are some samples of my favorite portrait outputs from Midjourney.
I also have included the AI prompts that I used to create them (unless otherwise noted, all of the examples’ prompts also included the word “portrait” so that the bot would focus on generating head-and-shoulders images).
Midjourney Examples of Portraits
All of the examples shown below are unedited outputs saved directly from the AI art creation tool (aka, I haven’t fixed them in Photoshop yet).
To see how slightly different prompts change the look of the same portrait subject, check out this blog post with more Midjourney examples.
Additional FAQs About Midjourney Portraits
What Makes Midjourney AI Such a Special Tool for Creating Portraits?
Midjourney AI is so awesome. It has the power to create something completely new out of just a few words. Even its goofs are pretty mind-blowing.
Whether you use it to explore portrait ideas or as a starting point for doing some overpainting, Midjourney AI is a mighty visualization app.
If you have trouble getting started, you’ll love that it makes the iteration steps go really fast.
Does Midjourney Steal Stock Art to Create New AI Art? Why Are There Watermarks on My Outputs? How Do You Stop the Midjourney Bot From Signing the IMages It Renders?
Sometimes, it looks like the Midjourney tool is copying and using stock photos and royalty-free images hosted by Getty or Shutterstock because there are artifacts on outputs. Adding something like –no watermarks or –no logos, etc., to your prompt copy is a good workaround to clean up your results and remove these distracting elements.
What Type of Art Does Midjourney AI Have the Easiest Time Generating?
Midjourney AI excels at creating images that aren’t realistic. It does a great job of purposefully drawing stuff that looks unnatural. If you search the #midjourney hashtag on Instagram or Twitter, you can see many great examples of unreal landscapes.
As such, it’s much harder to find really excellent outputs of Midjourney portraits. This is mainly because both users and the program itself seem to have a more challenging time getting stunning results for this type of art.
For instance, if you want to output “Bruce Willis as a robot,” it probably won’t be a problem. He’s bald and clean-shaven. The AI tool doesn’t have to worry about how to render hair, etc.
With a bit of fiddling and a few different outputs, you can probably knock it out of the park since you’re creating something that isn’t meant to look like anything that actually exists. However, if you want to make a realistic portrait of Zendaya, the program will likely struggle to generate something that believably resembles the multi-hyphenate star.
Are There Any Limits to Prompts Used With Midjourney?
You are only limited by your imagination. Still, there are a few “technical” restrictions.
For instance, you can’t use certain words. Midjourney limits some terms that have the potential to generate overly sexualized or extra horrific images.
All of this is spelled out, though, in the guidelines, so there are no surprises. Even with these minor restrictions in place, you can create pretty much anything your heart desires.
What Is Midjourney AI’s Greatest Weakness When Fabricating Human Portraits?
If you take a critical look at the program’s output, you can identify flaws, especially if you want to create more realistic portraits. For instance, image edges are often rough, and things often will meld together unpleasantly.
I can’t tell you how many times I’ve had to scrap an otherwise perfectly good Midjourney portrait because the program decided to do something like make a thumb grow out of the person’s neck.
What Is a Good Alternative Tool for Creating AI-Generated Portraits?
Many of the artists in the Midjourney community also rave about using DALLE-2 for creating portraits and other art; however, I haven’t used it. The pieces I’ve seen from it, though, make it look like a promising alternative to Midjourney.
I’m an Illustrator. Will Midjourney Steal My Job?
Much of the output from MidJourney AI won’t elicit that same feeling that art that was drawn by a real human will. Its pieces lack real depth and often have a slightly-detached and hollow look to them.
For example, during the mid-19th to early-20th century, many artists were nervous about the growing popularity of cameras. They worried photography might eclipse (or even replace) traditional types of art. However, today, as we all know, illustration and photography are both highly-regarded formats for art.
Inevitably, text-to-image AI portraits won’t ever wholly replace other methods. Instead, using computer tools will just be considered another way that artists can use for portraiture.
Why Should I Hire an Artist to Paint My Likeness if Midjourney AI Can Create Awesome Portraits?
It seems like it will still be quite some time before Midjourney AI and other AI bots have the potential to totally replace traditional portrait painters among consumers who want a piece of art for their home. That’s because these tools still require clients to describe what they really want with accuracy and clarity.
Midjourney isn’t going to ask you follow-up questions about composition nor follow all of your commands when it generates an output. Plus, even more simply, they don’t have the same limitless imagination that humans have.
Who Owns Portraits Generated With Midjourney AI?
Generally speaking, according to US and EU law, you can not copyright AI-generated art. According to the current terms of service for the tool, “All content generated … is owned by Midjourney.”
That said, currently, if you aren’t a paid member, you can use your portraits if you follow the Creative Commons Noncommercial 4.0 Attribution International License.
What’s the Deal with Paid Memberships?
If you work for a business that makes more than one million in gross revenue, you must buy a corporate membership to use the assets you generate for your company.
If you are a paid member, you can use your assets “without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, and sell copies of the assets.”
After Joining, Where Can You Learn More About Memberships and Other Rules?
For the latest and most complete information, go to the original source! You can do that by heading to the #rules channel on Discord. It may be templating to skip to making art, but you should review the latest terms of service first.
It covers restrictions for what you can do with any Midjourney portraits that you generate.
The Ethics of Midjourney Portraits and Other AI-GENERATED Art
I have several friends and family members who make their livelihood as photographers and illustrators. That’s why whether or not making art with AI tools is ethical is on my radar. Despite those connections, however, I’m not an ethicist or expert on the matter.
At this point, I use Midjourney and other AI tools solely for personal joy, so it’s not been an issue for me, but it is a matter that’s top of mind with many researchers and other experts.
If you’d like to learn more about this subject of debate, I recommend reading the Ada Lovelace Institute’s article covering different viewpoints as a starting point.
Still Have Questions?
Feel free to contact me on Discord. My username there is joyous#8616. I’d be happy to answer your questions (if I can) the next time I hop on to make some new art using Midjourney.
Alternatively, if you use Facebook, consider joining the Midjourney AI group. It’s a semi-active community of seasoned users and newbies alike where you can ask questions, troubleshoot problems, get ideas for new AI art, and share examples of your best work.
The only reason I’ve made progress with this cool tool is because of all of the kind, creative, and smart people in the group (and their willingness to pass along tips). Come join us!