Generate Expert Photography in Midjourney

Quick Links

• Use Keywords with Interplay
• Common Aspect Ratios
• Reference Images & Style Tuning
• Set Style to Raw
• Don’t Use “Realistic” Keywords
• Use the Latest Versions
• Follow Basic Prompting Advice
• Upscale your Images
• Use Generative Fill to Touch Up Abnormalities

Use Keywords with Interplay

This was a big learning curve for me when I first learned this.

Many other sites will tell you to be precise with your keywords, but what they really mean is to choose keywords with significant interplay.

Interplay: the way in which two or more things have an effect on each other.

Meaning, you want to choose keywords that have interplay—or a connection—with a huge amount of topics. This will give the AI so much more information and inspiration to connect styles in its database to create something unique.

For example, “a surreal photo” has a lot more interplay than just “Photo”. It tells the AI that the photo was taken professionally, so the camera used to take the photo was probably high caliber, and the photo will be a little more weird and artistic.

Check out the same prompt before and after I add the “surreal” keyword!

Interested in learning more about Surrealism in AI? Check out this Prompt Inspiration Deep Dive.

See Also

• AI prompt inspiration
• Photorealistic Photos in Dall-e

Use Common Photography Aspect Ratios

Most cameras take photos in 4:3, but can also use 16:9 and 3:2 rations (as well as their portrait counterparts).

Since Midjourney was trained on real photos with these dimensions, it does a better job at generating interesting compositions at these ratios.

4:3

Prompt: a bright photo of a fashion model posing for a hip hop poster in front of a graffiti wall during the day with small highlights of light blue, strong pose, female hip hop star –ar 4:3 –v 6.0

16:9

Prompt: a bright photo of a fashion model posing for a hip hop poster in front of a graffiti wall during the day with small highlights of light blue, strong pose, female hip hop star –ar 16:9 –v 6.0

3:2

Prompt: a bright photo of a fashion model posing for a hip hop poster in front of a graffiti wall during the day with small highlights of light blue, strong pose, female hip hop star –ar 3:2 –v 6.0

1:1

Prompt: a bright photo of a fashion model posing for a hip hop poster in front of a graffiti wall during the day with small highlights of light blue, strong pose, female hip hop star –ar 1:1 –v 6.0

Simply use the —ar parameter to the end of your prompt to ensure you are generating photos with realistic dimensions.

Use Reference Images & Style Tuning

Sometimes it can be hard to explain your style or composition to the AI, sometimes it’s easier to just show it!

You can use Midjourney’s Image Prompting to do exactly that! Simply paste the links to the images you’d like Midjourney to take as a reference at the beginning of your prompt and then watch the magic happen!

You can also combine Style Tuning with image prompting to take this further. Style Tuning is a Midjourney feature that allows you to train a subsection of Midjourney to a particular style. This way you can set your photography style using image prompts in your tune command, and then apply that to new images using the generated style code.

Set Style to Raw

By default, Midjourney applies its own set of styles to images. For the most part, this style is great! It truly just ensures the generated options are as perfect and beautiful as possible!

Perfect images, however, are a problem when it comes to generating photography. See, photos capture real life, and as we all know, real life is certainly not perfect.

This is why you should set your style parameter to raw when you’re generating photography. This way it allows a couple of those realistic imperfections into the generation photos. Things like more uneven skin tones, and less impossibly beautiful people.

Midjourney’s default style can also cause issues with photography images because the style might break the rules of reality. This isn’t an issue with paintings where physics and reality can be suspended to make room for creativity. However, since photos capture real life the rules of reality must be followed in order for the image to look realistic.

Check this out in the example below.

The wings in the first image look more like digital art than graffiti on a street wall, the shadows just don’t make sense with how light works in the real world. However, the wing art in the second image with the –style raw parameter looks much more realistic.

–Style Raw

Prompt:  a photo of a woman standing in front of graffiti art of wings –ar 4:3 –v 5.2 –style raw

Don’t Use “Realistic” in your Prompt

This one’s counterintuitive but if you are looking to generate a photographic image, don’t use “realistic”, “hyper-realistic”, “photo-realistic” or anything like that in your prompt. It will actually make your images less realistic!

Think of it this way: if you saw a photo of a cat, would you describe it as a hyper-realistic image of a cat? Or would you just describe it as a photo of a cat?

Now let’s go through this same exercise but this time we’re describing a really convincing digital rendering of a cat. How would you describe that? Probably would use the word realistic this time, hey?

Adding the keyword realistic insinuates that the photo is imitating realism like a painting or rendering would rather than capturing the real thing like is done in photography.

This might seem like a small difference, but understanding how we as humans naturally describe things is extremely important since our alt text descriptions are largely what have trained generative AI models such as Midjourney.

Just look what happens when I describe a photo of a cat, rather than a photo-realistic image of a cat.

A photo-realistic image of a cat

Prompt: a photo-realistic image of a cat on a busy city street with bright grafitti –ar 4:3 –v 6.0

Use the Latest Version of Midjourney

This is a simple one! Make sure you’re using the latest version of Midjourney in your settings.

As you might expect, the AI model gets better at photography with each version.

v 5.1

v 3

So if you want to generate the best photography possible, make sure you’re using the latest version.

As of the writing of this article, the latest released version is 5.2 but v6 is in its Alpha phase. This means v6 is still being tested by the public and it hasn’t been fully released yet so 5.2 is the default. However, you can test v6 yourself by setting it as your personal default in your settings or by using the —version parameter.

You should default to the latest released version, but it’s possible for you to set a different version as your standard. To check this, go into your Midjourney settings by typing /settings in your discord chat. Then ensure the latest version is highlighted in green.

Follow Basic Prompting Advice

Just because you are prompting for a specific medium doesn’t mean you should throw the basics out the window!

I go into the basics with many more visual examples in my Beginner’s Guide to Midjourney, but here’s a quick highlight of the types of things Midjourney is looking for you to describe. You by all means do not need to describe all of these areas in your prompt. These are just some categories that are meant to spark inspiration for your prompting. If you don’t described one of these areas, it will be left to Midjourney to fill in the blanks, which it is normally really good at.

Upscale your Final Images

This may feel like silly advice as you usually need to upscale an image to download it, but I’m talking about more than the original upscale. Using Midjourney’s x2 and x4 upscalers.

Upscale

Upscale (x2)

Upscale (x4)

Use Photoshop Generative Fill to Touch up AI Abnormalities

Although AI has made some pretty shocking progress in the last year, Midjourney is not perfect yet.

The AI still sometimes struggles to generate proper looking hands, or might add some incoherent noise in the background.

These oddities are usually so small you don’t even notice until you really inspect the upscale image.

So you might have generated the PERFECT image for your project… minus the fact that your models fingers are all connected to each other 😬

What do you do? Keep generating until you hit the AI jackpot? Or just pull the image into Photoshop and use Generative Fill to touch up the oddities.

If you do not have Photoshop, you can use Midjourney’s Vary (Region) function to fix small selections of your image. However, I find Photoshop’s generative fill to be much more effective at adjusting Midjourney’s small oddities.

To learn how to use Generative Fill, check out my beginner’s guide to AI in Photoshop!

Check Out My Photography Prompt Lists

Check out some of my other keyword inspiration lists. All prompts are tried and tested in Midjourney.