Why Use AI to Generate Photos for Your Project
Create More Inclusive Photographs
A lack of diversity is a well known fault of stock photography. Certain demographics such as the South Asian population, or anyone above the age of 40 are notoriously difficult to find.
With AI, you can make sure your project speaks directly to your specific demographic with its visuals.
You can generate professional looking photos with traditionally underrepresented people with a few simple prompts.
Jump ahead to check out my prompts on how to ensure there is more diversity in your AI photos.
Generate Completely Unique Photos
If you’re a designer who’s worked in the industry for a few years, you’ll likely recognize some of these photographs.
Hell, you’ve probably used them in a couple of your projects!
This is another issue with Stock Photos, everyone has access to the same library!
Naturally, there are some photos that get used more than others, and then your designs are no longer unique. Not to mention you have the risk of using the same photo as your client’s competition.
With AI you can generate photos that no one else has seen before. They are completely unique to you and your project.
Train Your Team to be Midjourney Masters!
Launch your team ahead of the curve with personalized Midjourney training.
Includes:
Prompts specific to your brand
Customized process documentation
3-hours of live workshops
How to Generate Stock Photos in Midjourney
This article assumes you already know the basics about how to generate photography in Midjourney. If you need a refresher, check out my article on how to generate realistic images in Midjouney.
1. Build Your Base Prompt
Without having a masters in art history, it can feel a little overwhelming creating your base prompt.
How exactly do you describe what you have pictured in your mind? How do you make it look like it was taken by a professional stock photographer?
With all that ambiguity, it can be hard to know where to start without wasting all your fast hours. That’s why I’ve put together a prompt template below to help you start generating stock photography.
If you need help with keyword inspiration or recommended parameters specific to stock photography, I have those below too.
2. Iterate on Your Prompt
Unlike what social media wants you to believe, you likely won’t generating the perfect image with your first prompt. It’s actually quite an iterative process.
You start with a base prompt and then add/remove from it as you see what Midjourney returns. You’ll have no idea which biases Midjourney has for your style, subject, or landscape until you generate that first image.
I usually plan for iterating on my prompt at least times before I am able to hone Midjourney in on what I’d like it to generate.
1st Iteration
Prompt: an asymmetrical stock photo of a happy mature korean man working, white with a little purple –ar 4:3
2nd Iteration
Prompt: an asymmetrical stock photo of a happy mature korean man working in an office, white with a little purple –ar 4:3
3rd Iteration
Prompt: an asymmetrical stock photo of a happy mature korean man working in a modern office, white with a little purple –ar 4:3
3. Upscale
Once you found a photo that’s at least 80% perfect, it’s time to upscale!
If you don’t know what upscaling is, check out my breakdown here.
You don’t need your initial image to be perfect. In fact, you’ll be faster if you choose images that are 80% perfect and then edit the imperfect parts out.
4. Edit Your Photo
Even if at first glance your photo looks perfect, make sure you zoom in and take a deeper look. Near all the photos you generate in Midjourney (or any AI really) will need to be edited.
Due the diffusion process behind how these models generate images, AI generated images will often have mistakes like weird hands or just noise that should be edited out.
Midjourney has a built in image editor, Vary (Region), that allows you to select areas of your image and prompt for a fix. You can’t select small areas in Midjourney, though. For smaller selections, you can use Photoshop’s generative fill.
Jump ahead for my guide on common areas that require editing in AI photos.
5. Add to your Project
Now your image is ready to be added to your project!
Midjourney & Firefly Tested Keywords
Monthly Inspo Right in Your Inbox
Sign-up and get style-specific keywords directly to your inbox every month.
Recommend Parameters
This article assumes you know the basics about what parameters are and how to use them. If you’d like to know more, check out my visual guide to parameters.
What parameters you use depends on the style you’d like to generate. For an example, while the –weird parameter works great when you are generating surrealist art where the nonsensical is encouraged, it wouldn’t work as well when generating Stock photos where everything needs to make sense.
The following are parameters that I often use while generating stock photography and the values I’d recommend using to get the best results.
–Aspect Ratio
–aspect or –ar
This parameter controls the dimensions of your images. By default, this is set to generate square images (1:1), however square is not usually the best aspect ratio for photography.
It is important to always remind ourselves how AI’s are trained. If we understand the world as they see it, we will be able to communicate with the AI better.
As a general rule, most photographs have an aspect ratio of 4:3. Knowing this, it is likely that Midjourney would have been trained on more 4:3 photographs than any other dimension. Therefore it is a lot better at generating 4:3 photographs.
Changing the aspect ratio from square to landscape also gives the AI more room for interesting compositions. Instead of just generating someone in the centre of a square image, Midjourney can play with more asymmetrical compositions.
It is important to keep composition in mind when selecting an aspect ratio. Midjourney tends to try and fill the space it has available to it. So if you are looking to get the full body of your subject into the photo, you are more likely to be successful with a vertical 3:4 aspect ratio than landscape or square dimensions. It’s the same concept as turning your camera on its side to capture someone standing up rather than trying to do that horizontally.
Last thing to consider is where you’d ultimately like to use this photo. For an example, if you’d like to use the image as an ad or as a hero image on a website, you’ll need to make sure there’s room for copy on the side. The 16:9 aspect ratio is really helpful here as it usually gives you more space for your headlines. If you’d like the image to be useful on phones or social media, the opposite is true: a 9:16 ratio gives you a tall photo that is perfect for mobile screens.
Original 1:1
a lifestyle photo of a fashionable diverse and mature woman wearing glasses outside
4:3
The most common aspect ratio in photography. Generates more interesting compositions.
a lifestyle photo of a fashionable diverse and mature woman wearing glasses outside –ar 4:3
16:9
Great for when you need a decent amount of copy room.
a lifestyle photo of a fashionable diverse and mature woman wearing glasses outside –ar 16:9
3:4
Vertical aspect ratios are great when you want to capture something tall like a person standing up, or a building.
a lifestyle photo of a fashionable diverse and mature woman wearing glasses outside –ar 3:4
9:16
The perfect size for phone screens.
a lifestyle photo of a fashionable diverse and mature woman wearing glasses outside –ar 9:16
–Repeat
–repeat or –r
Repeat is a pretty self-explanatory parameter. It repeatedly submits a prompt the number of times you specify.
I personally use this a lot after I’ve iterated on my prompt and found one that’s generating images I like. I then run a –repeat 3 so I can select from more options.
Max Repeats by Subscription
2-4
For basic subscribers.
2-10
For standard subscribers.
2-40
For mega subscribers.
–No
–no
The –no parameter is really helpful to get around Midjourney’s biases.
As I’ve discussed many times across my site, one of the most important things in prompt engineering is to keep in mind how these AIs were trained. Midjourney only understands the world based on the images it was trained on. Because of this, it has pretty ingrained biases.
For example, when you try and generate an image of a couple out biking, Midjourney often tries to generate them on the same bike. This often creates compositions that don’t make a lot of sense. I.e. floating handle bars, or two people sitting on one bike.
This is where you’d use the –no parameter. You can then list items you don’t want in your images (separated by commas) and Midjourney will do it’s best to avoid generating them.
This parameter is not perfect, but you will usually get at least one that follows your request. Plus, it is way more effective than typing “no shared bike” directly in your prompt; that will all but guarantee you’ll get people sharing a bike in your images.
Original
a stock photo of a diverse senior couple riding a bike outside –ar 4:3
–no shared bike
a stock photo of a diverse senior couple riding a bike outside –ar 4:3 –no shared bike
–Sref
–sref [URL]
Already generated a photo with the style you’d like other photo’s to copy? The –sref parameter identifies the colors, techniques, and composition of the the reference image, and then uses them in new generations.
This is especially helpful when you have to generate a lot of different images for the same project. Like for hero images on a website, a multi-asset ad campaign, or social media posts. This way your photos stay on brand without needing a lot of extra editing.
I have written another article that dives deeper into into the –sref parameter, but all you need to know for stock photos is to add the parameter to the end of your prompt, and then paste in your image’s link after that.
You can use multiple links, but I’ve found I have more consistency if I use the same single link over and over again. This way you know what the AI is keying on.
Original
a stock photo of a young professional working with a coworker, lofi filter
–sref
a young student listening to music with headphones by the window –sref https://s.mj.run/XS8-QJUiti4 –ar 4:3
–chaos
–chaos [number]
If used in low doses, the —chaos parameter is really great at diversifying the results you get.
When you are generating the first images of your project, you’re likely going to want to explore styles too. However, Midjourney tends to generate 4 images along the same note. A low valued –chaos can hello randomize those results a bit more.
I usually stick in the –chaos 15-20 range because anything lower than that doesn’t really have an impact and higher than that things get … well, weird!
Check out the results I get below. Without the chaos parameter, all my images are pretty similar; however, when I add –chaos 15, I get a little more variance in my photos. The second photo has a bit of a Mayfair filter on it, and the third photo has a bit of a brighter colour palette to it. The –chaos 20 photo grid takes that even further; every photo in that grid has something that makes it unique from the others. The –chaos 50 photos, in my opinion, get a little too unique. They lose their professional look and just look weird.
Original
a stock photo of a young south Asian professional walking down a busy street –ar 4:3
–chaos 15
a stock photo of a young south Asian professional walking down a busy street –ar 4:3 –chaos 15
–chaos 20
a stock photo of a young south Asian professional walking down a busy street –ar 4:3 –chaos 20
–chaos 50
a stock photo of a young south Asian professional walking down a busy street –ar 4:3 –chaos 50
Prompt Examples for Generating Stock Photos
Base Prompt:
“a stock photo of a young professional working with a coworker –ar 16:9”
Composition Prompts
Adding composition keywords to your prompt helps you generate more interesting photos. You are asking Midjourney to put special focus on how the photo is layed out rather than just defaulting to its usual centered composition.
Plus, it can be helpful to prompt for specific compositions to make sure your asset works for your project. Such as prompting for an asymmetrical composition so you make sure your asset has room for copy if you are using it for a website hero image.
Asymmetrical
An unsymmetrical, yet visually balence composition. Midjourney usually interprets this to mean placing the subject on one side of the image.
Prompt: an asymmetrical stock photo of a young professional working with a coworker, bright and modern
Leading Lines
This keyword tends to add a focal point to the image and then leads your eyes to the focal point. Its especially effective with outdoor photos.
Prompt: a stock photo of a young professional working with a coworker, bright and modern, leading lines composition
Symmetrical
This prompt tends to have a mirroring effect in Midjourney.
Prompt: a symmetrical stock photo of a young professional working with a coworker, bright and modern
Circular
The subject of the photo creates a circular shape.
Prompt: a circular stock photo of a young professional working with a coworker, bright and modern
Centred
Although this is often Midjourney’s default composition, calling it out in the prompt usually add more interesting deatils to balence the photo.
Prompt: a centred stock photo of a young professional working with a coworker, bright and modern
Monthly Prompt Inspiration
Learn new AI art styles with monthly style guides.
Bauhaus Prompts
Palette Knife Prompts
Fashion Photography Prompts
Inclusive Prompts
Midjourney is trained on images from the internet, so this, unfortunately, means that the model has inherited some our existing biases. While the company has made some strides to overcome some of those biases, the truth is Midjourney doesn’t usually generate diverse subjects automatically.
If you are looking to represent a more realistic look at the world, you are better to add diverse keywords to your prompts.
Diverse
This prompt tends to have a randomizing effect. It does tend to generate more of a African American demographic, but it will also generate other ethnicities and ages.
Prompt: a stock photo of a diverse young professional working with a coworker
Location Specific
By far, the best way to make sure your desired demographic is represented is to use location-specific prompts such as South Asian, Norwegian, African American, South American, ect.
Prompt: a stock photo of a South Asian young professional working with a coworker
Culturally Specific Clothing
Sometimes, biases can work in your favour.
You can describe clothing specific to certain cultures in order to generate images of a specific demographic.
Prompt: a stock photo of a young professional wearing a head scarf working with a coworker
Monthly Prompt Inspiration
Learn new AI art styles with monthly style guides.
Bauhaus Prompts
Palette Knife Prompts
Fashion Photography Prompts
Mood Prompts
How a photo makes you feel dictates its atmosphere, lighting, and style.
You can control a lot by describing the “mood” of a photo.
Gloomy
This prompt generates dark and dramatic photos.
Prompt: a gloomy stock photo of a young professional working with a coworker
Romantic
This prompt generates warm photos with engaging subjects.
Prompt: a romantic stock photo of a young professional working with a coworker
Enigmatic
This prompt creates mystiours photos with an impossible to figure out mood.
Prompt: an enigmatic stock photo of a young professional working with a coworker
Nostalgic
This prompt generates retro-looking photos.
Prompt: a nostalgic stock photo of a young professional working with a coworker
Emotion Prompts
Emotional keywords are similar to mood keywoords but they adjust the facial expression of the subject rather than the mood of the photo.
Happy
Prompt: a stock photo of a happy young professional working with a coworker
Sad
Prompt: a stock photo of a sad young professional working with a coworker
Confused
Prompt: a stock photo of a confused young professional working with a coworker
Frustrated
Prompt: a stock photo of a frustrated young professional working with a coworker
Filter Prompts
Sometimes it’s better to generate photos with the filter and styling you’d like to have in your final photo; the final style might influence the subjects and items in the photo.
Polaroid
This gives the photo a retro style as if it was taken by a polaroid camera. It often has the same white boarder polaroid cameras are famous for.
Prompt: an asymmetrical stock photo of a young professional working with a coworker, polaroid filter
Soft Focus
This keyword adds a warm and fuzzy filter to its images, often including a yellow hazey sun flair.
Prompt: an asymmetrical stock photo of a young professional working with a coworker, soft focus filter
Glitch
Glitch overlays digital-looking symbols and screens over the image adding a more technical look.
Prompt: an asymmetrical stock photo of a young professional working with a coworker, glitch filter
Lofi
The lofi keyword adds a purple-ish blue hue to all its images.
Prompt: an asymmetrical stock photo of a young professional working with a coworker, Lofi Filter
Monthly Prompt Inspiration
Learn new AI art styles with monthly style guides.
Bauhaus Prompts
Palette Knife Prompts
Fashion Photography Prompts
Keywords to Avoid
There are some keywords I’ve encountered that seem like normal words to describe photos that just don’t seem to work with Midjourney.
Laughing
This keyword often generates nightmare fuel images. People’s mouths seem a bit too big, the smile doesn’t always show in their eyes, and there’s almost always an uncanny valey aspect to the images.
Prompt: a stock photo of a laughing young professional working with a coworker
With a Highlight of [color]
While it is normal to describe a photo with small swatches of a specific colour a “highlight”, Midjourney often interprets this word as a lighting prompt.
Prompt: a stock photo of a laughing young professional working with a coworker, white with a highlight of red
Want More Photography Prompts?
Check Out My Other Articles!
Midjourney & Firefly Tested Keywords
Monthly Inspo Right in Your Inbox
Sign-up and get style-specific keywords directly to your inbox every month.
Common Areas You’ll Have to Edit in Midjourney Images
As you may well know, AIs don’t have a picture in mind and then create it. It slowly layers on pixels to an image until it creates something similar to what it’s seen in its training data. This means the AI doesn’t fully understand what a hand is, or what the blurred out objects in the background are. It just understands that other images have similar objects.
This process is what causes the weird hallucinations in AI images.
While image generators are only getting better every day, you’ll still be hard-pressed to find an AI-generated photo that doesn’t have some sort of hallucination you’ll want to edit out.
Here are some common areas you’ll want to keep an eye out for:
Hands
Teeth
Eyes
Artifacts
Clothing Symmetry
Train Your Team to be Midjourney Masters!
Launch your team ahead of the curve with personalized Midjourney training.
Includes:
Prompts specific to your brand
Customized process documentation
3-hours of live workshops
