Learning to prompt in MidJourney (or any other AI generator), is an artform unto itself, and one that keeps evolving as new features get added. But your best bet is just KISS – Keep It Super Simple.
I know…your first instinct is “LET ME PUT IN ALLLLLL THE POSSIBLE DETAILS”!! Maybe you’ve even thought to ask ChatGPT for a prompt that you can use.
The trouble is that ChatGPT is only trained on things up through 2021. So it knows nothing about MidJourney. There are folks out there training CGPT on the parameters MJ has in it’s FAQ, but still. What ChatGPT will give you could be a paragraph – or 6.
Let’s take an example here:
I’m going to ask Chat to give me a description of for creating a specific type of art. In this case, Patachitra – a type of eastern Indian scroll painting with intricate details, and mythological narratives and folktales. Let’s see what it gives me…
Could you give me a good prompt for MidJourney, an AI image generator, for some images that would resemble the eastern Indian Patachitra style of art?
Here’s what ChatGPT returned to me:
Generate a series of vibrant and intricate Patachitra-inspired images depicting traditional folk tales from eastern India. The artwork should feature bold lines, vibrant colors, and intricate detailing. The subjects can include mythological figures, deities, wildlife, and scenes from rural life. Each image should capture the essence of the Patachitra style while showcasing unique narratives and cultural elements.
This doesn’t look too bad, now does it? Not as long as it could be.
Ok. Let’s try another ChatGPT-style prompt for a photograph:
Please give me a detailed prompt for a professional close-up portrait-style photograph of a young woman with long red hair, outdoors, wearing a jade necklace?
And here’s what CGPT gave me back (note, I did ask for “detailed”, and that is certainly what I got!):
Capture a stunning close-up portrait of a young woman with long red hair, showcasing her natural beauty and vibrant personality. The setting should be outdoors, preferably in a location with soft natural lighting such as a garden or a park. The photograph should emphasize her facial features, especially her eyes, while highlighting the texture and color of her hair.
The woman should be dressed in a simple yet elegant attire that complements her red hair and enhances the overall aesthetic of the image. As specified, she should wear a beautiful jade necklace, adding a touch of sophistication and symbolism to the composition.
Pay attention to composition and framing, ensuring that the woman's face is the focal point of the photograph. Experiment with different angles and perspectives to capture her unique charm and individuality. Encourage her to express genuine emotions and capture those candid moments that reflect her personality.
Additionally, consider the background elements to create a visually pleasing and harmonious composition. Incorporate elements from the natural surroundings, such as trees, flowers, or foliage, to add depth and context to the photograph without overpowering the main subject.
Lastly, aim for a high-quality, professional result by using appropriate equipment and techniques. Focus on sharpness and clarity, while also maintaining a soft, flattering feel to the overall image. Pay attention to details like lighting, shadows, and the woman's pose, ensuring they contribute to the overall aesthetics of the portrait.
The final photograph should be a captivating and timeless close-up portrait, showcasing the woman's beauty, the allure of her red hair, and the elegance of the jade necklace she is wearing, while capturing her unique personality in a natural and authentic manner.
Oh boy. Not just one paragraph, but 6!!!. This is definitely not a KISS prompt for MidJourney – AT ALL.
BUT, let’s see what each of these different prompts produce.
Ok, first things first though. How do you even begin to create an image in MidJourney on the Discord server? That place is a mad-house, right? Yeah. I know. In the next post I’ll tell you how to create your own discord server with MidJourney, so you don’t have to wade through everyone else’s stuff to find your prompts.
But the basic idea is this…
When you get into one of their prompting groups on the server, you’ll see a box at the bottom that looks like this
In order to begin a prompt you need to start with
/imagine and then hit enter (or if you’re in the mobile app, click on “prompt”). This is what it should look like for you now
Your cursor should now be in the space after the word “prompt”. This is where you’ll enter your text.
Here’s the text ChatGPT gave us above, entered into the prompt area.
Hit enter, and now you wait just a few seconds (occasionally the servers might be way overloaded and it’ll take longer). In the MidJourney server, you may have to start madly scrolling to find your output.
This is what I got as a result of that first prompt. and HEY! Not too shabby! They look pretty promising, just at this small size already!
NOTE: the default aspect ratio for MJ is 1:1 (a square). We’ll talk more about how to change that a little further down
Let’s click on that grid of 4 to enlarge a bit and see a little more detail…
Well whaddya know? The prompt wasn’t really too bad afterall! These are actually quite lovely!
Ok. Now, looking back at the image before this, notice the U1-U4, and V1-4, and the rerun button? I’ll explain those to you now.
The U stands for “upscale” – in other words, a larger version of the image you choose. The numbers go left to right, top to bottom. So, the top left image 1, top right is 2, bottom left is 3, and bottom right is 4. So, clicking on U1 will generate a larger version of the top left image.
You’ll end up with an enlarged version like this
Now you can click on that enlargement, right click, and save it to your hard drive if you wish. But don’t despair of losing anything you’ve generated. It all lives over on your actual MidJourney.com account that you created when you joined the MJ Discord server (more on that in another post as well). Just know that, even if you have trouble finding your generated images in the crowded and busy MJ server, you will still be able to access them another way.
Now, there are a number of new buttons underneath your upscaled image, but we’ll look at those another time as well (I’m telling you, there is SO much you can do with with MidJourney, it will take post after post after post, and then some more, because they’re constantly adding new features.
But you’re probably asking, but what about the U numbers in the original grid. GOOD QUESTION! V stands for “variations”. Let’s see what happens when we click on V1, which corresponds to the image we just upscaled.
First, a Remix Prompt box will pop up. This gives you an opportunity to add or delete parts of the existing prompt. Or you can just leave it alone. In this case, I’m going to leave it alone and just click on “Submit”.
The following will appear on your screen until it actually starts generating the variations.
But that message will soon disappear, and you’ll begin to see your variations slowly materialize (there’ll be a percentage that shows as they load). And now we wind up with another 4-image grid of variants on that first image we upscaled.
And now we can just keep going down a rabbit hole with one single image. But you might want to try some other things too.
Ok, you’ve seen the general process for how this is done with a relatively short-ish prompt above. I’m not going to give you screenshots of all the steps this time, but I will show some results.
/imagine and hit enter. Now we’re just going to copy/paste alllllllll that text in that ChatGPT gave us.
Oooof…all that text is so long, it doesn’t even all appear in the final output.
Still, it’s a close-up portrait of a young woman with long red hair. She’s outside, the lighting is good, but there’s no jade necklace. It really seems like most of that prompt is useless drivel, right? Or, at least, MJ doesn’t seem to be quite getting all of exactly right.
So, let’s try an interesting (fairly) new (as of this writing) feature from MJ called /shorten.
Instead of typing
/imagine, you’re going to type
/shorten and hit enter for the prompt area to appear. Now, paste that huge prompt in and hit enter.
In the box that appears will be a button that says “Show Details”. Click on that first off. What’s the first thing you notice? Well, first of all, it only even read up to the first 128 words (yes, I counted them all there. I did not, however, count how many characters it read. Doesn’t matter though). And what are the 2 most important words it picked out of all that? “red”, and “long”.
You’ll also notice that it’s provided 5 shortened versions of the prompt. Let’s see what we get. Each one gets progressively shorter. And, to be honest, not all of these are terribly good. Really? “up, long red hair, garden”? Hmmmm. But let’s see.
Ok. The first one
Capture a stunning close-up portrait of a young woman with long red hair, showcasing, natural beauty and vibrant personality. The setting should be outdoors, preferably in a location with soft natural lighting such as a garden or a park. Ignoring the bad punctuation…40 words.
Those are just as lovely as the first one, I think.
Let’s see the results of the next shortest prompt:
Capture a stunning close-up portrait of a young woman with long red hair, showcasing, vibrant personality. The setting should be outdoors, location, natural, garden or a park. We’re down to 28 words now.
Capture a stunning close-up portrait of a young woman with long red hair, showcasing, vibrant, setting should, garden
Still coming out pretty great, with only 19 words – with some questionable punctuation again though.
Capture a stunning close-up, young woman with long red hair, garden. 11 words now.
Dang! Y’know…it’s still looking beautiful. Definitely more emphasis on the “garden” aspect here though.
And now for the 5th and shortest –
up, long red hair, garden. 4 whole words, down from the 6 paragraphs CGPT originally gave us.
Well, she’s definitely got red hair, there are plants around her, and it’s a photograph. Not really quite as successful, imho.
But let’s see if we can take what we’ve learned from the /shorten command, and see if we can reconstruct a logical prompt that’s not too short, and not too long. It’ll be the Goldilocks of prompts! I hope…
MidJourney tends to give most weight to words at the beginning. So let’s think about what the most important aspects are for this image:
For the most part, camera settings, camera types, f-stops, etc., mean nothing to MJ. By default, it assumes you’re prompting for a photograph, unless you tell it otherwise. There are a few camera settings that can be used though:
There may be a few others, but those are the main ones I’ve seen make any difference. Tilt shift might be kind of a hit or miss one though.
The important thing isn’t even just the length, but that you structure the prompt properly.
In any case, knowing that MJ’s default is to assume that you’re looking for a photograph, we don’t need to specify that. But we do want to make sure that important information is at the beginning. So, a good structure would be something like: ::important things::. ::a few additional details like color palette or environment or mood or clothing::. ::any other commands at the end::.
Learning the proper structure of a prompt is kind of like learning how to play a video game – once you learn the combinations, how to combine them, you eventually figure out how to control your opponent, with as little damage to yourself as possible. But until you learn those combinations that give you the upper hand, you keep getting killed and starting over. Or, sometimes you luck out and move on to the next level. But usually you have to figure out those critical combinations in order to consistently move through the game.
So, if we take our structure and our important image components and bash them together, what could we write?
This is what I think I’m going to start out with:
Formal headshot. Happy young woman, long red hair, jade necklace. Outdoor setting, soft natural lighting, casual clothing. Close-up portrait. --ar 8:10
Whoa, hold up! What’s that last bit?
That’s how we set the aspect ratio of the image. Most headshots are 8×10 (it’s also a good size for being able to print at home on your inkjet printer that doesn’t do edge-to-edge printing).
So…what am I going to end up with when I put in that prompt? Let’s see.
Remember, we start with
/imagine and ENTER so we can type in our prompt. And then we wait for the server to start rendering our first 4 choices.
I’m not hating these. But they’re more formal than I would have liked. Gee, I wonder why? My VERY FIRST WORD IS…Formal. BUT, we’ve got our young woman, we’ve got our long red hair, AND we’ve got a…necklace…I think at least one of those might be jade?. And we’ve even got the correct aspect ratio for a headshot.
But you know what? Headshots these days are often a lot less posed. People want to show their personalities a little more. Let’s see what we can do to up the personality in these images by re-doing the prompt a bit. Maybe make this more of a fashion shoot style. Let’s forget the jade necklace part. We’re going for bling now!
Happy young woman, long red hair, elaborate gold jewelry. Sitting on a bench in a flower garden. Profile view, un-posed. High fashion. Wide angle, tilt shift --ar 8:10
Many people will add in famous photographers to inform the style of photography. I try to avoid adding specific names if they’re current, living artists. But that’s just a personal preference. It can make a massive difference in the photography style you get back from MJ though.
Hmmmm…I’m a little underwhelmed. I think I actually preferred the results in our 1st /shorten version. So I’m going to go back and analyze that one to see how I can get something closer to that, but with good structure.
close-up portrait. young woman, red hair, gold jewelry, natural beauty, smiling, casual clothes. Outdoors, park, soft natural lighting. tilt-shift --ar 8:10
Yeah, I like these better. But do you notice a couple of problems in the two on the right? The top right one has some weird string hanging out of her shirt. And the bottom right one, the sleeve is a bit…wonky. Both of these could be fixed easily in Photoshop. But I’m going to see if I can fix them by just doing a re-roll. And then we’ll stop for today.
Yeah! That’s actually a lot better. I’m not seeing any glaring problems anywhere. But…I lied. I’m going to make two more subtle changes to the prompt, because I feel like her skin is just too airbrushed looking. I’d like it to be more natural.
So…first, I’m going to take out the “tilt shift” part of the prompt. It tends to cause a bit of blur, and that could be contributing to some of it. But I’m also going to add another parameter at the end
--style raw. This basically removes all the default MJ photo style levels, and just gives me a raw photo output (ish). I like how these came out much better, but I wish there still wasn’t quite so much air brushing. I may play with it a bit more after this, just for kicks, and report back when I figure it out.
So, for the record, my final prompt for our red-headed gal today is:
close-up portrait, young woman, red hair, gold jewelry, natural beauty, smiling, casual clothes. Outdoors, park, soft natural lighting. --ar 8:10 --style raw
And these are the 4 results upscaled.
This was a long one, y’all. I hope you hung in there with me. Like I said…learning to prompt effectively in MidJourney is a bit of an artform all unto itself. It’s a combination of logic, imagination, and technical know-how. Believe me…I do not know everything about prompting for MidJourney. I doubt anyone knows everything. But I enjoy sharing what I do know, and what I continue to learn along the way. We’re going to cover a lot together in the next installment as well, including…greebles!!
If you feel like seeing everything I generate, from the good, to the bad, to the awful, to the weird, and everything in between, feel free to check out my MidJourney page here.
And finally, the full-color version of the header image, using a modified version of the Patachitra prompt thusly:
vibrant and intricate Patachitra images, depicting traditional folk tales from eastern India. bold lines, vibrant colors, intricate detailing, mythological figures, deities, wildlife, and scenes from rural life. --ar 16:9