📱

Get Our Mobile App

Take your business learning on the go!

Download on the App StoreGet it on Google Play

Prompt Engineering Tips for Midjourney & BETTER AI Art!

Wade McMaster - Creator Impact18:56

Transcription

Improve your prompt engineering with Mid Journey or just about any AI art platform using the following tips.

The first tip is to use positive words in order to focus on what we want to see and not what we don't want to see. Because if I type into Mid Journey "a farm with no dogs," all of a sudden there's a dog because Mid Journey just reads the word "dog" and decides to include it in the image. But change the prompt to "a farm with cows, pigs, and various animals," and suddenly you get an image which has those animals in the scene. So focus on what it is you want to see.

And if that doesn't work, then you can try using negative words in your prompt. And it works in a pretty simple way. You type up your prompt and then at the end, you type `--no dogs` to remove dogs from the prompt. I hit enter, we get images of farms and no dogs.

The next tip is to put your priority words at the start of your prompt, as Mid Journey and other AI art platforms will prioritize what is at the start of the prompt in your image. So if I type in "a massive evil bull on a nice farm," it'll put the bull front and center and prioritize it in the image. But if I switch it to "a nice farm with a massive evil bull," then it tends to reframe the image to prioritize the farm a lot more. And this can be used also for styles or anything that you want to prioritize in your image.

Another example of that is "a massive evil bull on a nice farm, minimalist logo." Popping the style at the end. Or I move to the front with "a minimalist logo of a massive evil bull on a nice farm," and you can see there's a little bit more of a focused difference.

Next, I would consider defining color and lighting in your prompt in order to take a little bit more control over how your image looks. This time I have "a masterpiece painting of a massive evil bull and a nice farm," and I add "pastel colors" to get a pastel color palette. Or if I want something a little bit different, "neon colors." Add something in to find exact colors such as "blue and green color palette," or I can just say "sepia colors" for that old-fashioned look.

Or in regard to lighting, I can say things such as "low contrast" or "high contrast" if I want to affect the image. And these images were made with the exact same seed, so you can see the exact result in the difference in those prompts. But you can also just mention things like "dark low lighting" to get a nice dark image, or "very sunny lighting" to get something that's very sunlit and bright. Or that in between, you can go for "golden hour lighting." There are a lot of different options you can choose when describing how your image is lit and the colors being used.

Next is to take advantage of word weight and multi-prompts, specifically in Mid Journey, in order to fine-tune the various areas of your prompt. The way this works is we add two colons after a word and then a number to score it, one being the default. Any higher than that will increase the priority of that word.

So if I take this image of "a massive evil bull in a nice farm" without any scoring, it gives us an idea of where we started. But for reference, I've added a one next to the bull. This time, but the farm, I've added `::2` to get it to emphasize the farm more than the bull. But we switch it back and we add, say, a score of three to the bull and one to the farm. You can see how much more that the actual bull has been referenced here.

But for a little bit of fun, I've made the word "evil" a score of eight, "a bull" a score of four, and everything else default. And you can see just how it's drastically altered the overall look of that image by really focusing on the word "evil" and then the word "bull," and the rest comes in at a much lower priority.

Also, you can remove the number and use what's called multi-prompts. A good example of this is: I have an image of "an outside hot dog," which is a picture of a hot dog. But what if I just want "outside hot" and a picture of "a dog," not the actual food? I type in `outside hot:: dog`. And those two colons will separate that word. And now Mid Journey reads it as "outside hot" as one part and "dog" as the other. And of course, we get a picture of a dog.

The next tip is to read, learn about art, photography, or anything visual. This will allow you to expand your vocabulary and get better at describing what it is that you want to see. So actually studying this, getting involved in it, will help you communicate better with Mid Journey or any AI platform when you are prompted.

And that leads into the sponsor of today's video, Skillshare. I'm always looking out for more tips to improve my YouTube channel, so I'm going to check out this YouTube Success course by Marcus Brownlee, which covers a whole bunch of topics to allow me to improve and grow on YouTube. But there are a vast number of extra classes like this available in just about all fields of creativity.

Now, chances are you've heard of Skillshare before. Basically, it's an online learning platform where you can go in and find certain courses around what it is you want to learn, take a bunch of classes, and improve your skill set around those topics. Now, Skillshare is the largest online learning community for creatives. Now, they have made some upgrades. They've included smarter class categories, and they actually cover a lot on their platform, including AI, but also other skills like I mentioned, art, or you want how to use Photoshop, any visual creativity style, any kind of creative business endeavor you have, like graphic design, web design. Skillshare has classes that you can check out, and they're very intuitive and easy to follow. Skillshare has thousands of classes on that platform, or just about any topic you can think of.

So if you want to check it out, the first 500 people who use the link in the description below will receive a one-month free trial, so you can check it out for yourself. Take your time and see what it's all about. I want to thank Skillshare for sponsoring this video, and now let's get back to it.

My next tip is not to make your prompts too long, if possible. Try to keep them short and dense and focus only on the important aspects of the image. This could be difficult because the longer your prompt is, we tend to get as much detail in there as possible. But by doing this, some of the things we put into our prompt can distract from the more important elements. So we need to really define what is really important and keep our prompts as short as possible.

Here's an example of an image which has a rather long prompt. If I click on this prompt, it will sharpen the prompt bar. If I want to add details in here, it may be difficult to, and it may be competing with a whole bunch of the information in here. And we need to really focus on what's important if we want to have a really efficient prompt that's easy to actually adjust.

So we can see here, we've got the word "image." Straight away, we're generating an image, so we don't need to worry about that. And we have "composed of brown, green, and olive color palette." If we just hit comma, "earthy tones," we can cover a lot more ground with that phrase as opposed to spelling it all out.

Now I'm going to get back to the start. It says here, "Create an artwork of a small monkey in the stall of modern video game built on Unreal Engine." Now, also, it has "his small monkey," and then "the monkey is standing." So we can completely rephrase this to shorten it. So we can just say, because a video game is built on Unreal Engine, we don't need to worry about the word "video game." We don't need to be so descriptive. We can just say "an Unreal Engine 3D render of a monkey." And we say, "see the monkey standing in a forest environment with trees, beautiful nature, and bright blue sunny sky in the background." First of all, the sky is always going to be in the background, so we can get rid of that. We can just say "a sunny sky." And we can delete and just simply add a comma. I'm not going to worry about the monkey standing, because the monkey is likely going to be standing because of the type of images that Mid Journey is going to produce with the subject of a monkey. So I can just say "a monkey, forest." A forest generally has trees. Nature generally has trees. I don't think we need to worry about nature, because forests are composed mostly of nature. We can look at these things and see what is important and what isn't. So I can basically get rid of "environment with trees, beautiful." I can get rid of "nature."

Now we have this really short: "I don't even have to have an." I have "Unreal Engine 3D render of a monkey, forest, sunny sky, high contrast, earthy tones." So we've got this much shorter prompt now, which is going to be easier for us to add details to if we need to make any adjustments. But let's hit enter and see what image we get.

So you see, we've actually got more earthy tones than we do here. So it's actually followed the color palette a bit more closely, not competing with as many things. Has a bright sunny sky. He's in a forest. The only thing I can notice is that he's sitting. So again, I can come up here, make the adjustment of "a monkey standing," hit enter. We have more images where the monkey is standing, although he's still sitting here. But we managed to sort of steer Mid Journey in the right direction. I'd also like to point out the original prompt described the monkey standing, and he is sitting in all of these images. So by not competing with this lengthy prompt and really focusing on the important aspects, I'm able to get a little bit more control over my image by being more direct and simple.

However, another great way of doing this is to head into Discord. Once in Discord, I can type in `/shorten`, paste in my prompt, hit enter. You'll notice it shortens, focuses on the important tokens. So it has shortened the prompt, and I'm able to sort of use these as a way of trying to remove the distracting words from my phrase. So I can try it. It removes the least from the first and the most from the last. So I can create these images. And now, starting from the first and going through to the last, you can see the image we get here is not too bad. The next one is actually even a little bit better, a little less of the earthy tones. Even this very short prompt looks pretty good here. This even shorter prompt is starting to get a little bit away from the earthy tones. And this one has removed the forest. So simply using "shorten" and picking the prompt that produces the best result with the shortest prompt is a great way to take the important part, add it to your image, and then you can make some adjustments that will have a bit more of an impact on your image.

A powerful tip that also keeps your prompt shorter is to define distinct styles of your image. It's a pretty flexible tip, but there are things like common art styles like "a watercolor painting," or you can try something a little bit different like "an ink dripping drawing." But if you want a photo, you can add something in like "DSLR photo," or even go for something a little more cinematic like "a sci-fi blockbuster movie screen cap." If you want to try something like a bit of anime, there's always "anime 1990 film screen cap." Using Mid Journey, you can do something a little bit inventive. Also, I have here "an ice sculpture," or I can even try "any style that comes to mind." Have a bit of a play. Or materials. I have a full list of styles you can try in my style video playlist. I'll check that out and I'll pop a link in the description below.

And another step in that direction is my next tip, which is to cover a lot of ground by using artists' names. As certain artists have very particular looks that could be similar to what you're trying to create. Some artists you can try: I'm a huge fan of H.R. Giger, he always seems to produce some pretty unique style of art. But you can try someone more famous, someone like Picasso, if you want that real Picasso style. Or even "in the style of Salvador Dalí," if you want something a little bit stranger looking. Or even "in the style of Leonardo da Vinci" to get something that's a little bit more like him, a little more of a golden tinge to it. You can try the style of Hans Bellmer, or something else like a combined style, the combined style of Hans Bellmer and Picasso to get something unique, or the combined styles of H.R. Giger and Leonardo da Vinci. You can combine artists or try one artist at a time to get very unique, very outstanding styles.

Another tip is to get inspiration from looking at other people's prompts in order to get an idea of different ways that you can word your prompt and different phrases and words you can use. Now, there's a few different places out there you can go, like there's freeflow.ai, or there's social media. I've got a full video on this, I'll leave in the description below. But also heading to the explore page of Mid Journey, you can scroll until you find something that looks pretty cool.

So this image here looks a bit different. I can use this prompt by coming down to "use prompt," change a few things around. I can also just have a look at the prompt and see if there's anything in there I want to try. I can simply click back, come in here again. Again, we've got a nice big long prompt. I could try and shorten that, look for different phrases in there. I can go through and see how people have styled their prompt. So this image here, we can see we get this very nice sort of fisheye lens effect. You can see here, "fisheye lens effect." So by going in, looking at their prompts, you can find this kind of information. Also, things like "film grain" to add a little bit of grain to that photo. Go through, look for some inspiration for prompts by using Mid Journey's explore page or some of the links I'm going to leave in the description below.

My next tip is to take an existing prompt, split it apart into shorter phrases by using permutations to see what results you get from each phrase. Now, I'm back on the Mid Journey explore page and I found this creepy-looking image here. It also has a prompt with a few different phrases in there. So I can click "use prompt" down here to add it to the top. I'll make my subject a dog. I type in "dog." But if I want to try some of these different styles first, I'm going to remove the parameters and I'm going to use permutations by adding a curly bracket at the start and a curly bracket at the end, and anything separated by a comma, it will essentially switch out these word for word. So I've got "antique photography in a gothic city," "in the 1900s," "dark fantasy," "WTF," "Dark Soul style mythical beast," "weirdness has no boundaries," "in dire world." I think that's supposed to be "dire world." So now if I release that, I'll get a prompt for "dog antique photography in the gothic city," or "dog in the 1900s," or "dog dark fantasy." And the reason I do this is because I want to see what each phrase, what impact each phrase has on the image, and I can start to sort of collect the phrases that work best for me. I hit enter, come back to create.

So by splitting that prompt apart, I have "dog antique photography in gothic city," and that's a pretty unique result. Scrolling up, we have "dog in the 1900s," very old-fashioned style photography. "Dog dark fantasy" has given us a very misty sort of appearance. "Dog WTF" has also given us a bit of a fun result. "Dark Soul style mythical beasts" gives us something a little bit different again, while "weirdness has no boundaries" also gives us some strange images. And of course, "dog dire world," where you get some more unique images. So this is a great way to discover star phrases you can use in your prompt.

Another tip is to describe the angle or the framing of your image, so that way you can again get a bit more control over how the image is presented. So in this image, you can see we've used the term "high angle" to get a high angle photo of the bull. And with "low angle," we get something that's taken from a bit of a lower angle, sort of looking up at the bull. You can also frame it up, so we have a nice "landscape shot" where it has a landscape in the background showing more of the full bull. Or you can get nice and close with a nice with a "closeup shot" of the bull by mentioning closeup. Or you can even say something like "far away landscape shot" to get a bit further back from the bull. Or going back to the angle, we can add "side angle" to show the side of the bull as well. This is another way you can control the perspective of your images.

Often, the first image is not enough. So my next tip is to consider always rerolling and remixing your image to sort of go down that rabbit hole in order to find what it is you're after. Now, I cover both the web and Discord here, but it's pretty simple that if I really want to try and get another version of this, like it's close but not quite what I want, I come down here, hover over a rerun and click that, and it reruns that prompt and gives me some new results. However, if I don't make some subtle alterations to the prompt, again, I just click the prompt, come over to my prompt here, and type in what I want: "muted blue tones." Submit. And I produced another set of images, this time with that blue muted tone that I added into the prompt.

In Discord, if I want to reroll, I can simply click this reroll button here, and it does the same function as hitting rerun on the website. But there's something else I recommend doing there: is typing in `/settings`, hitting enter, turning on remix mode. Mine's highlighted green, so it's already on. So now when I come up, I click the same button, and I get a chance to alter the prompt. So I can either leave it the same, and I always leave this turned on if I want to use the same prompt. I just hit submit and keep it the same. Or I can change to "muted purple tones," hit submit, and I then get my revised prompt and a new set of images. And that is a great way to go through and make little adjustments to your image to try and get something closer to what you're after.

My next tip is to describe a certain mood in your prompt in order to get a certain look or feel for the image you're creating. "Depressing mood" is a great example, as it tends to dull things down and make the image a little bit more muted. Let me move over to "excited." We get a lot of energy, a lot of movement, bright, bright scene. "Gloomy and spooky," another good example of making things spooky, almost a haunted feeling. And "quiet," similar again, not really much movement, like a quiet morning. We switch to "loud," you get a lot of energy, a lot of things up in the air. But when you switch up to something like "powerful energy," you get this really fiery, intense energy to the image. But bring it back down with "serene stillness," and you get this nice, serene, still-looking image. Try something different, we have "vibrant party energy." It shows you how using different moods and descriptions can really change the overall look of your images and is incredibly powerful.

My next tip is to use solid prompt formulas that will help you achieve a better image, but also help speed things up, as it gives you the information you need. All you need to do is fill in the blanks and create your image from there. Here is an example prompt formula you could try to use: "Art style of a subject in a locational scene, mood." So switch out these various square bracket items for what you want to see in your image.

The first example is "a 3D render of a gray star alien in a spaceship, gloomy and spooky." But change the location: "We have a 3D render of a gray style alien in an office, gloomy and spooky." We get something a little bit different. Change the mood to "powerful energy," and we get something vastly different again with the alien in the office. But if I change the alien to a dolphin: "We have a 3D render of a dolphin in an office, powerful energy." Or we can pop him back in a spaceship: "Have a 3D render of a dolphin in a spaceship, powerful energy." If this is starting to look a bit absurd, I'll switch it to "a detailed pen sketch of a dolphin in an office, gloomy and spooky." And again, make another change to a spaceship, and we get a different style yet again. This is a nice, easy way to switch out bits and pieces and experiment with prompts to see what kind of results that you can get.

If you're looking for more prompt formulas, I'll pop a video on the screen right now. You can check that out. That will get you started with some prompt formulas in Mid Journey. Otherwise, that's a video for today, guys. Thanks for watching. Hope you see you again next time. Have a great day.