Midjourney can generate beautiful images from a simple text prompt that consists of only a few words. But if you want to see what this text-to-image model is really capable of, there are certain techniques that you should use. In this article, I'm going to talk about how to write good prompts for Midjourney.
While I do encourage you to explore what type of results you'll get from writing various raw text prompts, you should also focus on learning how to write advanced Midjourney prompts. This may sound complicated, but it really isn't.
It would be silly of me if I told you that I can teach you everything there is to know about writing good prompts for Midjourney. In fact, I advise you not to listen to anyone claiming to have this knowledge.
We're still very much in the early days of our understanding of how to interact with text-to-image models. Our collective knowledge increases each day, which means new ideas, techniques, and tricks will get discovered often.
You will learn a lot about writing prompts simply by interacting with the model and seeing what other people are writing in the Midjourney Discord server. With that being said, I have some neat tricks that I want to show you that will significantly improve your Midjourney prompts.
How to Use Weights in Midjourney
As soon as you join the Midjourney Discord server, you will notice a lot of other users writing advanced prompts. All users start off writing raw text prompts, but after a while, they start adding various symbols, parameters, and numbers. It may seem like it would be extremely difficult to reach that level - but it's really not.
Let's start talking about advanced prompts by utilizing text weights. You can think of weights as numerical values that you use to emphasize a certain component in the image you want to be generated. While this numerical value can range from -10,000 to 10,000, it's more efficient to use numbers in a far smaller range.
The default weight of each word in your prompt is 1, which means that you'll have to assign any other value if you want to emphasize or de-emphasize a certain component.
Let me give you an example so that you better understand text weights in Midjourney. I'll start by instructing the model to generate an image of a cookie jar, which you can see below.
If I want to use text weights, I have to add two colon symbols and a number to specify the value of the weight. If I want to emphasize cookies in the image, my prompt would look like this:
/imagine cookie::2 jar
And here's the result I would get.
As you can see, the cookies are quite big. In two of the four generated images, there is only one huge cookie in the jar. I'll now alter the prompt a bit so that it looks like this:
/imagine cookie:: jar::3
In this prompt, I made sure that the weight of the cookie is 1, while the text weight of the jar is 3. This put a lot more emphasis on the jar, as you can see in the image below.
This is a pretty simple example, but you can use text weights in more complex prompts. You can assign weights to multiple different components in the image you want to be generated.
Choose Your Style
When you spend more time experimenting with Midjourney, random ideas will start to pop up in your head. Sometimes, you won't be able to find the right words in order to get desired results. In case you want to take your AI-generated art to the next level, you will have to research different art styles.
It's important to figure out which words will produce a certain result. You might be able to pick up some cool ideas from other people on the Discord server, but it's also vital to do a bit of research online.
Start by familiarizing yourself with some of the most prominent artists of the previous century. Getting to know what type of art styles people like Salvador Dali, Andy Warhol, Piet Mondrian, Pablo Picasso, and Francis Bacon used will help you find the right words to describe what you want generated.
Let's say that you wanted to generate a drawing of a human heart. The most basic, raw prompt that you can write is:
/imagine a drawing of a human heart
Using this prompt would give you the following results.
The results you get might be completely different than what you'd wanted to see. If that's the case, it's because you didn't specify the art style. Let's see how different the results can be if we specify the style. I'll now enter the following prompt.
/imagine a drawing of a human heart in the style of Leonardo da Vinci, schematic, old drawing, intricate details, small details, accurate, forgotten drawing, mystic
Here is what Midjourney generated from the prompt above.
This is much different than the first four images generated. These results were also more consistent because I specified the art style and sprinkled in a bunch of relevant keywords in the prompt.
I want to go on a small tangent here to talk about the importance of keywords when you're interacting with text-to-image models.
Generating the perfect image often has a lot to do with the keywords you use, and you should feel free to use as many as you want as long as it makes sense in your prompt. I'll give you an example.
/imagine a neopunk style image of a futuristic city with huge skyscrapers that emit purple and blue light, intricate, wondrous, awe, future, virtual reality, ready player one, digital, insanely detailed --ar 16:9 --v 5
Here's the result we got from this prompt.
In essence, what's crucial here is that you get familiar with the terms you need to create a good image. If you're into creating photorealistic art with Midjourney, you should learn about different camera lenses. This will make it a lot easier for you to generate the art that you're looking for.
I can't overstate how important it is to find the right words for every prompt. You should pay attention to the details - everything from the overall art style to the type of lighting you want in the image.
Exclude Elements from the Image
One of the things I like a lot about Midjourney is that it gives you the option to easily exclude elements from an image. This is really helpful when you don't want to risk the model generating something you don't want in your images.
Let's say that you want to create an image of a burger. In many cases, you will instruct the model to create an image like this and it will automatically generate French fries right next to the burger. Even though you didn't even mention French fries, the data it was trained on made it notice countless instances in which French fries were right next to a burger.
Text-to-image models like Midjourney don't understand what they are generating. They can't tell whether the user wants French fries next to the burger, which is why you have the option to exclude any elements from your images.
All you have to write is "--no" and mention the element you want to exclude right next to it. Let me show you what I mean with an example:
/imagine photorealistic image of a small burger on a wooden board --no french fries --v 5
When I do this, I essentially make sure that the model will generate only what I want it to. See the result of this prompt below:
It's difficult to predict what the model will generate when you enter a certain prompt. It might take a few attempts before you figure out what you want to exclude from the image. I'll show you a very simple example.
I entered a prompt instructing Midjourney to generate a bowl of fruit. Here are four images that it generated from this prompt.
There are grapes in every single one of these fruit bowls. The fruit bowl is one of the most important exercises for an artist since they use it to work on their lighting techniques and compositional skills. It's also an exercise that is meant to help artists learn how to bring items to life when they're painting.
As you can imagine, there are literally countless images of fruit bowls on the internet. If I type in "a bowl of fruit" on Google, I'll see grapes show up in many of the images. In Midjourney, I can instruct the model to exclude this fruit from the image.
This time, I wrote the following prompt:
/imagine a bowl of fruit --no grapes --v 5
Here's the result.
The first fruit bowl image was generated by the fourth version of Midjourney. This one was generated by the fifth version. It's interesting that this one is way more photorealistic even though I didn't instruct the model to generate the image this way. But as you can see, there are no grapes in the bowls.
How To Write Good Midjourney Prompts
As you can see, it's really not that difficult to start writing better Midjourney prompts. Simple tips and tricks like the ones I mentioned in this article are enough to make you better than most people who've used this text-to-image model so far. To summarize, you want to try and:
- Use weights to emphasize a certain keyword in your prompt
- Learn about different art styles and then instruct Midjourney to generate images in a certain style
- Make sure to exclude elements that you don't want to see in the generated images.
I recommend that you spend as much time as you can researching different art styles and keywords that you can use to improve your prompts. Using the right words will have a significant effect on the output you get.
It's crucial that you find adjectives that you can use for different prompts. If you've already spent a lot of time in the Midjourney Discord server, you've likely seen people writing prompts that contain more than 50 words.
It takes time to be able to quickly come up with a long prompt that makes sense and produces beautiful results, but following the advice in this article will help you get there.
