Visual Notebook
Chapter 3
Prompting Notes
Table Of Contents
Chapter 1: Beginner Notes
Chapter 2: Discord Notes
Chapter 3: Prompting Notes
- Workflow → Some methods I use to create with AI.
└ Additive, Poetic, Image Input - Prompt Writing → Things to think about when designing prompts.
└ Main Idea - Descriptors → Terms to define your image
└ Descriptor, Medium, Camera, Lighting, Color
└ Artist, Movie, Game
└ Weights (::) Separators (,) - Image Prompts → Use your own images as prompts.
└ Uploading, Image Weight, Image Blending - Prompt Batching → Send a bunch of prompt through at once.
└ Repeat (multiples), Permutations (variations of terms)
Chapter 4: Parameter Notes
Workflow
I have two main workflows, and I really enjoy both ways of working. Basically, in one I spend a lot of time developing the prompt and testing it piece by piece until I get what I want, and in the other I just let the AI take the wheel.
Additive
Using an additive prompting technique is a great way to create and learn at the same time. Get a basic idea in your head, but don’t be dead set on what you imagine – let AI guide you as you guide it. How about a floating city in space! Type in your main idea to set the scene then run it. See what it looks like, then think about where you might want to steer it. Add a new descriptor like “art nouveau style” and run it. Then add more, or remove a term if you didn’t like how the image evolved. Keep testing to see where the idea is going… This method really teaches you what different terms do (or don’t do) for the final output, then you can help others with your knowledge!
(click the prompt to copy it!)
/imagine prompt: a floating city in space --v 5b
/imagine prompt: art nouveau style of a floating city in space --v 5b
/imagine prompt: art nouveau style of a floating city in space sacred geometry --v 5b
/imagine prompt: art nouveau style of a floating city in space sacred geometry winter --v 5b
Once you get a really awesome scene going, you can try getting more complex with it. Are ideas going away? Add weights (learn more about that HERE). Really want a certain style? Use a photo URL in the prompt and try to get it to come through (learn how HERE). Perhaps you like an image but want to remix it to Niji (learn about remix HERE). Maybe you start to care less about the original idea because you got a new idea by working with the AI. Go with it!
Poetic
As an artist, many times I look at AI as a form of television. It’s entertainment for me, but I am in charge, and the controller has a few billion buttons! I enjoy typing random things, but many times the results are amazing and might inspire bigger ideas. Try simple juxtapositions, lyrics from a Grateful Dead song, or pull a quote from the Bhagavad Gita and run a few variations. Let’s see what these look like in Version 4.
(click the prompt to copy it!)
/imagine prompt: half man half clock --v 4
/imagine prompt: in and out the window like a moth before a flame --v 4
/imagine prompt: I am become death, destroyer of worlds --v 4
Late one night (possibly after a nice glass of Scotch, or something) I typed in a bunch of complete gibberish into MidJourney, not even real words, and got some of my favorite images to date. When you let the AI go crazy, you get some really interesting images. Also, a fun tip is to add a lot of stylize (like –stylize 1000 on V4) or chaos (–chaos 100) just to see where things go. Have fun!
(click the prompt to copy it!)
/imagine prompt: the fleeboing zappony fliddling skakamuck googinsapiet totomuckas --v 4 --chaos 100 --stylize 1000
Image Input
Prompting with text is fun, but images in Version 4 can take you to a whole new level. You can upload a photo of yourself to try to get a selfie standing on the moon. Or think functionally. One idea I had was to upload a single letter of the alphabet at a time and create some complex fonts with each. Lastly, feed a MidJourney image back into itself (simply copy its URL link into the prompt), combine a few, or add some new words to evolve it. Read all about images in the Image Prompts section.
(click the prompt to copy it!)
/imagine prompt: IMAGEURL a man in an astronaut suit standing on the moon, dark sky and stars in the background --v 4
/imagine prompt: IMAGEURL 3D rendering of the letter S with textured metal surface:: top down photography of a motherboard, wires, transistors, backlit glowing led colored lights, computer fans, electronics, against a black background::3 --v 4
/imagine prompt: IMAGEURL1 IMAGEURL2 --v 4
Prompt Writing
With the onset of artificial intelligence, visual artists and creatives are now magicians of words, and we craft our imagery with letters, syllables, and descriptors. Okay, that kinda romanticized things a bit, but it’s kinda true?
When creating Prompter, I broke down prompts into categories of descriptive terms, like medium, color, movie, or lighting. Below are visual examples of each category to show the basics of how you can alter and add to a prompt (and soon I will be adding links to visual examples of hundreds of terms, and Version 4 as well).
Main Idea
AI is so powerful that it is up to us to imagine interesting scenarios. You got this! Be creative. Be funny. Be deep. Be yourself. You can borrow descriptive parts of prompts from other people, but the main idea should really be yours. Have you been feeling a certain way and want to portray it in the language of imagery? Did you have a dream last night of something so surreal only art could show it? Perhaps you just want to make someone laugh by combining silly things. Maybe it was something from childhood, or something from the future! Anyway, the main idea is the special sauce of your prompt, and everything else just helps form and shape it.
(click the prompt to copy it!)
/imagine prompt: a friendly fox --v 4
/imagine prompt: a treehouse --v 4
/imagine prompt: an evil monster --v 4
Descriptors
When I created Prompter, I broke down the language of a prompt into various descriptive themes, and then organized these themes into a database of terms. Below are the eight categories and what they do. Even if you don’t use my Prompter tool, this is a great way to think about building out the details of a good prompt.
Descriptor
This section is for other descriptive terms that don’t really fit into the other categories, just additional ways to alter your prompt. I include two descriptor sections in Prompter, but you can multi-select, so pick what is useful to you. You can make things more complex (fractal / insanely detailed), speak about time (1950s style / Low Poly), add elements (ice / fire), and so much more. Obviously, the list can be endless, but I provided enough in my database to get you inspired.
(click the prompt to copy it!)
/imagine prompt: a friendly fox, fractal --v 4
/imagine prompt: a treehouse, low poly --v 4
/imagine prompt: an evil monster, ice --v 4
Medium
The medium defines the main style of the output and is usually the most important descriptor. I find that placing it before the main idea of your prompt usually works best. This term can be as vague as an art movement (relief carving / psychedelic art style), a specific type of flat output (ballpoint pen drawing / blueprint), dimensional output (cross-stitch / relief carving), or even types of photography if you are a camera geek like me (calotype photograph / expired polaroid).
(click the prompt to copy it!)
/imagine prompt: relief carving of a friendly fox --v 4
/imagine prompt: blueprint of a treehouse --v 4
/imagine prompt: expired polaroid of an evil monster --v 4
Camera
The camera sets your point of view, your lens, and how the scene is spatially presented. You can go from one extreme (microscopy / macro lens) to another extreme (tilt-shift lens / satellite imagery), define focal traits (bokeh / motion blur), or even attempt to bring out the views in photography (50mm lens / 80mm lens).
Keep in mind that some terms don’t always work well. If you prompt an 80mm lens, you might get a nice image of a Fuji 80mm lens, lol. Also, microscopy can be cool, but a fox makes no sense, so instead, you get a fox kind of interacting with a microscope-like thingy, like below.
(click the prompt to copy it!)
/imagine prompt: a friendly fox, microscopy --v 4
/imagine prompt: a treehouse, satellite imagery --v 4
/imagine prompt: an evil monster, motion blur --v 4
Lighting
The lighting descriptor defines if there is a special aspect to light and shadow. You can use terms to describe really dramatic forms of light (contre-jour lighting / film noir lighting), a time of day (golden hour sunlight / sunset lighting), or certain forms of light (infrared light / laser light).
(click the prompt to copy it!)
/imagine prompt: a friendly fox, contre-jour lighting --v 4
/imagine prompt: a treehouse, sunset lighting --v 4
/imagine prompt: an evil monster, laser light --v 4
Color
Sometimes when your prompt doesn’t contain a particular mood, the colors default to a blend of cool and warm, cyan and reds. If you really want a specific color style, mention it. You can be blunt (green / dark purple), describe types of color (neon color / vibrant color), computer displays (1-bit grayscale / RGB), or even really strange mixes (diffraction pattern color / tie-dye color).
(click the prompt to copy it!)
/imagine prompt: a friendly fox, dark purple --v 4
/imagine prompt: a treehouse, neon color --v 4
/imagine prompt: an evil monster, 1-bit grayscale --v 4
Artist
As an artist myself, I find it really fun to see what different scenes look like by other artists. it is entertaining to mimic or combine. There is so much art out there, everything from Art Nouveau (Gustav Klimt / Antoni Gaudí), over to Post-Impressionism (Vincent van Gogh / Henri Rousseau), or even some photographers (Edward Burtynsky / William Henry Fox Talbot).
(click the prompt to copy it!)
/imagine prompt: a friendly fox by Jackson Pollock --v 4
/imagine prompt: a treehouse by Joan Miró --v 4
/imagine prompt: an evil monster by William Henry Fox Talbot --v 4
Movie
Adding a movie or TV show (I thought of calling it Film, but let’s just go with Movie), if the AI is trained on enough of the style, can really alter colors and elements. For example, some have a lot of drama (Dune / Spirited Away / Stranger Things) but others don’t work so well. I don’t use these often, but sometimes it’s fun. The new Version 4 works really well, and I’ll have updated images soon.
(click the prompt to copy it!)
/imagine prompt: a friendly fox, in the style of the film Dune --v 4
/imagine prompt: a treehouse, in the style of the film Spirited Away --v 4
/imagine prompt: an evil monster, in the style of the film Stranger Things --v 4
Game
I’m like half-gamer(ish). I grew up with Atari, NES, GameBoy, but a few (Quake / Doom / Zelda) were my favorites, many years ago. I would design levels and host deathmatches in Quake 1. In college I got everyone hooked up for LAN battles on Quake 3. It was epic at the time (a time when phones could hardly take pictures and streaming a movie was unheard of). Anyway, I enjoy some VR from time to time now, but gaming will always come in waves for me. Combining video games with AI is really fun (and even better V4/V5 images).
(click the prompt to copy it!)
/imagine prompt: a friendly fox, in the style of the video game Quake --v 4
/imagine prompt: a treehouse, in the style of the video game Tetris --v 4
/imagine prompt: an evil monster, in the style of the video game Zelda --v 4
Weights
Double Colon (::) → A double colon is a hard break to split up parts, also known as a multi-prompt, or also used to add weight (see below). This is actually a super-useful tool to break up one thing from another in your scene.
Weight (::2) → You can add weight, more focus or importance, to ideas using the double-colon right after them (blue::2), the higher the weight, the more the AI will focus on that idea, and negative weights (::-0.5) will get less or no focus (the same as using the –no command).
Weights can go from -10000 to 10000, but don’t think of super high numbers as meaning more. Weights are relative to each other and the whole, like adding the numbers together and dividing up parts. So if you use (blue::2 green::5) it will make exactly the same image as (blue::2000 green::5000). Lastly, you cannot weight a prompt negative, meaning you will get an error with (red::2 cyan::-10) because the overall added weight is negative. Get it? Got it? Good. See examples below:
(click the prompt to copy it!)
/imagine prompt: clock::5 clouds::2 green::-0.5 --v 4
(clock gets most focus, less on clouds, and likely create images that are not green)
/imagine prompt: clock, clouds::2 green::-0.5 --v 4
(commas do not separate an ideas, so in this case clock and cloud are weighted 2)
/imagine prompt: clock:: clouds::2 green::-0.5 --v 4
(here I added a :: to split up clock and clouds, now clock is 1 and clouds is 2)
/imagine prompt: URL1.jpg ::2 URL1.jpg ::2 --v 3
(lastly, you can use :: to weight multiple images differently, but remember to add a space before the :: separator)
Separators
This is a strange one. A lot of people format their prompts in various ways, using commas to separate ideas, periods, plus signs, and so on. I have done some testing and found that it does indeed change the image, but not really in a meaningful way. Basically, things are similar but different, but not in some logical way where you can say “this is the best format”.
Comma (,) → A comma is a soft break in an idea, however, the only time they are really needed is in the rare case of terms that could have two meanings. For example “a bright light house on the corner” will give you an ocean lighthouse whereas “a bright light, house on the corner” gives you a house and bright lights.
Other than that kind a scenario, I have tested things with and without commas and saw little logical difference, so maybe this is just a good way to appease our brains so things look more organized. So I say, use commas in a prompt if you want. Whatever makes you happy.
One last thing to mention, and something I just learned. Though commas rarely do much of anything, they actually have an interesting use on the MidJourney website. If you have a prompt with some commas in it, click the image in your gallery, then click any of the parts between commas and it’ll show you “previews” of other images with those words in them. Kind of a cool feature. If useful, this might make you think about using commas even if they don’t do much to the image itself.
Single Dash (-) → A single dash is said to be a good way to connect two or more terms into one idea, similar and opposite as the comma situation above. So while we all know “grandfather clock” is a big fancy clock, AI might think an old man, and a clock (maybe not, but just as an example). So you could type “grandfather-clock” to tie the words into one idea. That said, some test
Others → Sometimes I see others using periods, brackets, parenthesis, and other odd punctuations in their prompts. Again, I don’t see any logical reason to add more punctuation when it doesn’t really steer the image. I assume the AI pretty much ignores them anyway. Maybe skip that unless you like to hit your keyboard more than you need to. I usually run tests when I see someone doing (((this))) to see if it does anything, and usually, it doesn’t (that’s more for Stable Diffusion anyway, but I see people do it). Maybe they are just getting their AIs mixed up… Either way, below are a bunch of tests using the same seed so you can compare grids for yourself:
(click the prompt to copy it!)
/imagine prompt: a giant castle red sky green grass fog --v 4
Image Prompts
In Version 3 and below, MidJourney uses image input for style, color, and so on – so don’t expect to use a photograph of something and get a similar image in return. That said, in Version 4 and up are amazingly powerful for image prompts! Though hit or miss, you can do a lot with them. I sometimes have luck with faces (like my self-portraits below) however I really find image prompts to be useful to define the look of a landscape or altering the design of the output.
Uploading
Since image prompts use URL links, the easiest way to use an image is to send a JPG to the Bot, then click the image, click Open Original, then copy that URL into your prompt (you can skip the last step by right-clicking Open Original and copying the URL). Another method, if the image is visible in your char window, you can simply type /imagine then click and drag the photo into the prompt area and the URL will appear. If you really want parts of the image to be carried over try describing the specific details in text after the URL.
You can also feed a MidJourney image you make back into an image prompt. Do the same thing, click on the MidJourney image, Open Original, then copy that URL.
Lastly, you can add image weight to make an image more or less important than the text prompt. This is discussed in greater detail in the Parameters chapter.
(click the prompt to copy it – and yes, this URL is my photo of my, so have fun lol)
/imagine prompt: https://s.mj.run/IypGveyzp9I a character from the muppets --v 4
Image Blending
/blend → You can use the /imagine command, but the easiest way to blend multiple images is the /blend command. Simply type this and you can drag and drop images into the prompt. No need to upload and copy links!
Image Blending can be fun, but also functional. For example, you can mix two photographs to combine elements (like the example below with mountains and boats) or you can attempt to alter the material or style of an image by combining it with a specific look (like changing the white rabbit below to metal or wood).
Prompt Batching
Standard and Pro Plan folks get the addition of some nice batching features, so keep in mind these will not work for Basic Plans. They are currently limited to 40 prompts, so anything past that will be ignored. This feature also only works in fast mode, so it WILL eat up your GPU hours if you do it a lot.
–repeat → This parameter can be placed at the end of your prompt with a number in order to run the same prompt multiple times. If you want to run “a cute puppy” five times simply type “a cute puppy –repeat 5” and sit back and enjoy the doggo show.
Permutations → This is a fancy sounding word that basically means you can make multiple prompts with slight variations. So if you wanted to try the same prompt with a man and woman, simply add {man, woman} and it will run two prompts, swapping the terms out. You can also use parameters such as –stylize or –ar to change an aspect ratio like {–ar 3:2, –ar 16:9, –ar 2:1}.
You can also use {} in side other {} to multiply things further. As a visual example, let’s say we wanted to compare some 2D designs with 3D, but also make the colors different as well. I could add this {2d vector icon {cyan, magenta, yellow}, 3d low poly model {red, green, blue}} to a prompt to run three 2D prompts and three 3D prompts all at once. Below is our prompt, the resulting six prompts, and examples from MidJourney V5:
(click the prompt to copy it!)
/imagine prompt: cityscape {2d vector icon {cyan, magenta, yellow}, 3d low poly model {red, green, blue}} black background --ar 5:4
cityscape 2d vector icon cyan black background –ar 5:4 –v 5
cityscape 2d vector icon magenta black background –ar 5:4 –v 5
cityscape 2d vector icon yellow black background –ar 5:4 –v 5
cityscape 3d low poly model red black background –ar 5:4 –v 5
cityscape 3d low poly model green black background –ar 5:4 –v 5
cityscape 3d low poly model blue black background –ar 5:4 –v 5