Skip to content
I. Text Generation with ChatGPTi) Overview & Prerequisitesii) Generating Marketing Copyiii) Advanced Prompting: Meta-Promptingiv) Coding Assistant II. Image Generation with Midjourneyi) Overview & Prerequisitesii) Before You Prompt: --tile and --v3 iii) Creating Your First Printsiv) Advanced Prompting: --chaos and Multi-Promptingv) Creating Brand Imageryvi) Advanced Prompting: Deriving /describe
It takes a first stab at revising your initial prompt,It gives you suggestions, which you can type in answers to for the AI to incorporate them, andQuestions that enhance the precision of your prompt, and cover any information you might have left out that could be of help!
If you didn’t like any of the images, you can “re-roll” the prompt, and generate an entirely new set of images.If you like one, press its corresponding U button—this indicates you’d like to “upscale” that image. After a moment, a full-scale version of your selected image will appear. Right-click on it to find the “Save Image” button.If you liked the general vibe of one or more of the images, but want to see slight variations of it, click on the photo’s corresponding V button to create variants of that image.
Your prompt inputs aren’t limited to just text—you can upload existing images as prompts in Midjourney! This is tremendously powerful for us visual thinkers.Images generated using the default version of Midjourney are 1024px x 1024px in size. This typically works for images, but for printing on fabric, you might need higher resolutions to avoid pixelation showing up on your finished garment. There are several free upscalers available online that do a decent job. is free, doesn’t require an account, and will typically increase your image size by about a third (any larger will normally cause an error due to running afoul of its file size limits).You might be wondering what --ar 9:16 means, the strange gibberish at the tail end of the example prompt above. This is an example of a parameter, indicated by the two preceding dashes. Parameters allow you give Midjourney additional information as you craft your prompts. --ar is a parameter that specifies the aspect ratio you want your image to be—I wanted the aspect ratio of the prompt above to be 9:16, which is a common full-screen mobile aspect ratio. Try inverting this parameter yourself—instead of --ar 9:16, give it --ar 16:9. This will give you desktop-computer sized images instead!Another fun parameter is --chaos. This parameter accepts any value between zero and one-hundred, with the default being zero. Higher chaos values will generate higher variations among the four individual images in each job. For example, if you were to try the same prompt above, but crank the chaos up to eleven:
The first sentence supplies a setting (”A hot summer day in Ibiza”), and optionally a specific model/posture/action,In the next sentence, the garment is described in as much detail as possible (”She is wearing a blouse with shoulder tie, in the style of turquoise and white, densely textured or haptic surface, playful patterns, patterns, horizontal stripes, gossamer fabrics.”),Finally, some semi-technical photography details are specified (”Editorial for Vogue Paris, vintage polaroid”).
Use image prompts! Either directly in your prompt, or by analyzing them in Midjourney’s own words via the /describe command (see the following section for more on this technique).Don’t shy away from very short prompts! For example, throwing everything I know out the window, I’m going to specify a Major Arcana, and one of Christopher Alexander’s :
Midjourney
Table of Contents
I. Text Generation With ChatGPT
i) Overview & Prerequisites
ii) Generating Marketing Copy
iii) Advanced Prompting: Meta-Prompting
Before continuing on with this example, give ChatGPT/Midjourney a shot entirely on your own. Don’t overthink it, just be direct in describing what you’d like to generate from the model, and go along with your first instinct. If you can’t think of anything, try `Advertising for a Cinderella-themed
Spend fifteen minutes or so doing this, I’ll be waiting right here for you!
You’re back? Great! How were your results? How close were they to what you were picturing in your head?
“Freehand” prompting the models can be a valuable experience—it gives you a bit of intuition as to how the model does, or doesn’t, follow your instructions. But it can also be a bit unwieldy, having to guess how the model will interpret certain aspects of your prompts. You’re probably getting a lot close-but-no-cigar results, and feel like you’re under- or overshooting the level of specificity you need to provide.
Thankfully, there’s a solution...fight fire with fire!
No, seriously! Use the AI to help you better use the AI. This strategy goes by several names ( ), but for our purposes we’ll call it meta-prompting. It’s the difference between trying to drive through an unfamiliar city on your own, sans-map, with only a vague idea of where your destination is, and having Siri give you audible turn-by-turn directions.
A good meta-prompt will accept your initial prompt, and walk you through several subsequent iterations of the same prompt, arriving at a final prompt that is much more descriptive and precise, in the exact way the AI needs it to be.
For example (follow along if you’d like, it’s a great way to learn!), the meta-prompt below does three things:
Now, type in a prompt you attempted when you were still working on your own. In my case, I was trying to generate code for a webpage interaction:
The AI will return with a edited version of your prompt, slightly more optimized than your own, make suggestions, and ask you clarifying questions:
You can go through this loop as many times as you like! You certainly don’t have to answer every question or address every suggestion, and the model is smart enough you can answer everything stream-of-consciousness, no formatting required.
After a handful of iterations, you’ll likely reach a point of diminishing returns. But if you compare where we started, two gray boxes up, and a prompt that has gone through several iterations, the difference is night and day:
iv) Coding Assistant
II. Image Generation With Midjourney
i) Overview & Prerequisites
/imagine
/imagine is a command you use to tell the Midjourney bot you want to generate an image. There are other commands too, including /blend, /describe, and /settings, that we’ll go into later!
But for now, we’ll stick with /imagine. As soon as you type it, you’ll see a dropdown menu appear, and you’ll be prompted to indicate it’s a command by pressing return on your keyboard—press it! Now, you’ll be able to enter a prompt to specify what you want kind of images you want the bot to generate for you! Try adding the example prompt below to the text box, once you’ve engaged the /imagine command:
/imagine A detailed fashion sketch of a gorgeous evening gown by Charles James, charcoal on paper, c. 1950s --ar 9:16
Give it a few seconds to generate, and...you’ve just generated your first set of images! I’ve added my results form the prompt below; your’s likely look very similar.
Midjourney always generates four different images for each prompt you give it, to give you different options to select from. Once you have these results, you can do one or more of the following:
And that’s all you need to get started with Midjourney! Before we move on to our first example, here are a few more helpful general things you should know:
/imagine A detailed fashion sketch of a gorgeous evening gown by Charles James, charcoal on paper, c. 1950s --ar 9:16 --chaos 100
You’ll notice there’s much greater variation among the images—we have one with color now, a photorealistic example, and two different flavors of impressionism. --chaos is a useful tool if you don’t know exactly what you’re looking for, since you can sample a much higher variety of styles and concepts.
Now, you’re ready to get started with generating images in Midjourney! Let your imagination run wild, or continue along with this guide to learn how to create beautiful editorial images, and even repeating prints for your garments!
ii) Creating Brand Imagery
All of the images above were generated with Midjourney! Seriously! You can give a few of them a shot on your own below, they should produce something very close (or even better!) to the examples above:
/imagine A secluded Moroccan Villa Dior editorial for Vogue Italy, shot by Mario Sorrenti. She's wearing floral shorts and a model shirt in a soft pink shade in the style of striped dark beige and bronze. Medium shot, Fujifilm Pro 160C film still, 4k
/imagine A lazy summer day in Paris Dior editorial for Vogue Paris, shot by Tim Walker. A gorgeous 20-year-old woman with long blonde hair, slightly blush cheeks, and laughing. She's wearing a stunning floral midi dress ivory cs0101o 3 l cdi045df 00, in the style of light beige and amber. 35 mm Kodachrome film still --ar 16:9
/imagine A hot summer day in Ibiza. She is wearing a blouse with shoulder tie, in the style of turquoise and white, densely textured or haptic surface, playful patterns, patterns, horizontal stripes, gossamer fabrics. Editorial for Vogue Paris, vintage polaroid, --ar 9:16
You’ll notice that all of these prompts follow a similar formula that includes three things:
This formula is incredibly flexible, and really only limited to your creativity! In the rest of this section, I’ll give you a ton of ideas for each you can mad-libs together into your own creations. There are a ton of out there as well, if you want to go really deep into promptcraft!
Some ideas for settings:
- A bright spring morning in TriBeCa.
- A lazy summer day in Paris.
- A secluded Moroccan Villa.
- A hot summer day in Ibiza.
- A night out in Berlin.
Some examples of garment descriptions:
- A stunning floral midi dress in ivory.
- She's wearing floral shorts and a model shirt in a soft pink shade in the style of striped dark beige and bronze.
- An amazing pair of embroidered lace jeans.
- A Beatnik-style dress from the 1960s
- An opulent, Hollywood-regency style minidress with tassels.
(Note: A big need fashion designers have is to see their particular garment reflected in the images they generate, down to the exact print and construction. This level of specificity is quite difficult and not exactly Midjourney’s forte, but we’re working on some other solutions for this! Follow along as we figure it out in the last section of this doc.)
(Note: In the meantime, try using the /describe command on images of your exact garment, and then use the terms Midjourney uses to describe the image in your prompt! Some of the examples I gave have some strange wording—this is because I pulled them from Midjourney’s own vocabulary! For more explanation on this, skip down to the “Advanced Prompting” section below.)
And finally, some photography directions:
Lighting/Lighting Techniques
- Natural Lighting
- Moonlight
- Studio Lighting
- Overcast/Cloudy Days
- Off-Camera Flash
- Hard Fill Lighting
Angles
- Closeup Shot
- Low-Angle Shot
- Medium Shot
- Wide-Angle shot
- Birds-Eye View
Film Types
- Agfa Vista 400, 4k
- 35 mm Kodachrome film still, grainy
- Fujifilm Pro 160C film still, 4k
- Daguerreotype
- Polaroid Vintage 4k
Fashion Photographers/Publications
- Yves Saint Laurent editorial for Vogue Paris, shot by Tim Walker
- Schiaparelli editorial for Vogue China, shot by Zhang Jingna
- Dior editorial for Vogue Italy, shot by Mario Sorrenti
Again, Madlibs is a valid and productive strategy here. Choosing at random...
/imagine A lazy summer day in Paris. She's wearing a Beatnik-style dress from the 1960s. Off-Camera Flash, Closeup Shot, Polaroid Vintage 4k, Dior editorial for Vogue Italy, shot by Mario Sorrenti --ar 9:16 --chaos 25
Or, for something with a little more drama...
/imagine A night out in Berlin. She's wearing an opulent, Hollywood-regency style minidress with tassels. Off-Camera Flash, Birds-Eye View, 35 mm Kodachrome film still, grainy, Schiaparelli editorial for Vogue China, shot by Zhang Jingna --ar 16:9 --chaos 50
iii) Generating Prints
Regardless of what kind of prints you want to produce, there are three basic things you should know about print generation in Midjourney: the --tile command, the difference between the two relevant Midjourney versions, --v3 and --v5, and the power of /remix.
First, the --tile command is what allows you to generate seamless repeat images. You always want to append this to your prompts! Otherwise, you’ll end up spending hours in Photoshop, painstakingly editing the edges of your image to line up perfectly. In ye olde days of Midjourney print generation, there was no --tile command! So take it from me, your old, grizzled guide, who wasted many, many hours of her life fixing imperfect repeats by hand: use --tile!
Simply add it to the end of your prompt, taking care to add the double-hyphen (also known as a “flag”) before the word.
/imagine retro 60's circle print --tile
Above, you can see the same prompt without the --tile command, and then with.
(That second result brings up other thing worth noting—if your prompt for a print is to generic, you’re very likely to encounter all manner of watermarks! So be creative!)
Next, versions. Midjourney is actively under development, and typically every three or four months, they’ll put out a new version of the image generation algorithm. Currently, they’re on their fifth version. This newest version is a huge step up from previous versions in both amount of detail it can render, and complexity of prompts it can accept as inputs. Version five, notated as --v5 in prompts, is automatically selected for you when you first start out in Midjourney—you don’t have to change any settings to use it!
You should definitely start off generating your prints with --v5. But, if you’re using Midjourney for print creation, I highly recommend you try out the older, third version (--v3). It has a kind of “lo-fi” warmth and organic-ness that’s very difficult to get out of --v5. It might not work for every aesthetic, but it’s magic for those it does work for, and it’s certainly an alternate option worth exploring if you’re new to Midjourney!
To switch to --3, type the /settings command, and select --v3 from the available options. You’ll stay in --v3 until you chose to switch back to --v5. You can use the same exact prompts in --v3, and just like --v5 it accepts the --tile command, so your prints will repeat seamlessly (though it’s worth noting --v3 repeats sometimes feature slight imperfections, so be sure to double-check before you create a garment with your print!).
With --tile and the differences between --v3 and --v5 in mind, you’re ready to give print creation a try! I’ll give you a first prompt we can walk through together:
/imagine Art Deco pink repeating pattern, charcoal on paper, Paris c. 1930s, Surrealism --tile --chaos 50
As with the photographic generations in the previous section, Madlibs is as good a strategy as any! My formula is usually something like the above—specify a style and a color, and add the words “repeating pattern” for good measure, specify a medium, add a setting, and optionally add further artistic or historical terms. And again, always use --tile. And a lil’ --chaos 50 is never not a bad idea!
You should also try running this first prompt in both V3 and V5. Here are my favorites from my initial four results, from V3 and then V5:
One handy bit of advice is to use /remix whenever you “re-roll” your prompt or ask for variations on a generated image. Normally, when you “re-roll,” the same prompt is run again immediately, ut if you use the /settings command and engage “Remix Mode” by tapping on its button...
...now, whenever you hit “re-roll” or ask for variations, a textbox will pop up in the center of the screen, and you’ll be able to edit your prompt, while still maintaining memory of the existing prompt.
This is very useful for “jamming” with a single prompt. Take the pink V3 image above (the first one). Let’s ask for variations on it, but with a specific direction: I want to use the general pink and black charcoal vibe the image already has, but I want to see it rendered as Byzantine-style crosses, so I’ll replace most of the prompt with that:
Knowing which parts of the prompt that when altered will take you towards where you want to go, versus which not to alter to keep it harmonious with what you liked in the original generation, is a bit of a trial-and-error thing, so just try things! (But do take care to ensure --tile is still in there!)
Here are my results that I liked:
Not 100% what I was imagining (no crosses), but in this case, I personally find these two prints to be even better!
One last bit of guidance: printmaking is a lot more personal to each creator’s style—so definitely come up with your own formula that works for you! I won’t give any example descriptors as I did with photography generation for this same reason, because your stylistic preferences and descriptions are what will make this magical for you!
When in doubt, try one or more of the following to progress towards your own aesthetic:
/imagine The Chariot, Local Symmetries --tile --chaos 100
I used V5, and tiled it 3x3 in the second image so you could see the repeat a little better. Super cute!
\
🚧 iv) Advanced Prompting (TBD) 🚧
The following tips are a little more difficult than just appending a simple command like --tile to your prompt. They require some trial-and-error to build up an intuition for what’s happening inside Midjourney’s algorithm, what it likes and what it doesn’t.
a) Seeds and Additive Prompting
Consider the following two prompts:
b) Weights and --no
c) Multi :: prompting
works very well with images and images + text, too!
d) {Permutations}
e) Deriving from /describe
Given that it has seen literally every image on the Internet, Midjourney has a lot of broad knowledge of regularities across imagery and aesthetics generally. This knowledge is normally hard to add into your prompts, though, because these connections are implicit in the model. We don’t know exactly how a large AI model like Midjourney thinks, so the magic keywords it uses to “understand” these aesthetic regularities are not terms humans would be likely to come up with on their own.
That is, until /describe was released. /describe is a command that allows Midjourney to look at an image you supply, and express how it perceives that image in its own terms.
- floral midi dress ivory cs0101o 3 l cdi045df 00
- in the style of turquoise and white, densely textured or haptic surface
🚧 v) Very Advanced Prompting (TBD) 🚧
Want to print your doc?
This is not the way.
This is not the way.
Try clicking the ⋯ next to your doc name or using a keyboard shortcut (
CtrlP
) instead.