Midjourney
Last updated
Last updated
We're assuming you alread have registered to Midjourney's Discord service.
The first thing you need to do is choose a channel. You can work in your own private channel or in a public channel (newbies or general) you’re assigned to. There are pros and cons to each.
Private channel: you can get here on Discord by clicking on ‘private messages’ on the top left of your Discord app window, and then clicking ‘Midjourney Bot’.
Pros: you can work without distractions and keep track of your iterations. Cons: you are on your own, without inspiration or sense of community.
Public channel: you are assigned either ‘newbie’ or a bit later ‘member’. You can roam other channels to your heart’s content, but you can only generate images in your assigned channel.
Pros: you’ll have TONS of inspiration from others’ work and prompts. Cons: the channels are HECTIC and just keeping track of your work is a job.
PRO TIP: You can combine the best of both worlds by working on Midjourney Bot’s private message channel on your desktop app, while keeping the public Newbies # or Members # channel open in your browser. This way you can work in peace, while still getting inspired.
After choosing your channel, creation always starts with the ‘/imagine’ prompt. No exceptions. This is fastest done by just typing ‘/’ and pressing tab, as it is the default command.
After that, prompts and parameters are how you talk to Midjourney, so it’s critical to be fluent in both. Elements closer to the start get more importance and prominence, so a typical flow is as follows:
The first part of your prompt is always the URL to your image, IF you’re using one.
You absolutely do not need one, but it’s handy in inspiring composition, e.g. rough distance, number, angle and posture of characters or other objects.
NOTE: Midjourney is not initializing an image based on your image, nor is it modifying or editing it or building on top of it. Instead, it will always generate a completely new image. It will simply use your image as inspiration, just like it uses your text prompt as inspiration.
You’ll get the image URL for your own image best by: - Uploading an image to channel via the ‘+’ on the left side of the input field - Right clicking on the image posted to the channel to open a menu - Clicking ‘copy link’ to copy the link URL - Pasting the resulting URL right after ‘/imagine prompt’
NOTE: you can use more than one image, but for starters I would keep it to one image max.
NOTE: with an image the default weight it will have in inspiring the artwork is 0.25. We can change this, and we’ll also set custom weights to parts of the prompt, so remember the concept of weight.
The second part of your prompt is always the main idea itself: what do you want to create?
The options here are limited only by your imagination. Through extensive trial and error and based on some baseline lessons from photography I’ve developed the below useful practices.
Objects and Subjects
A good distinction to understand from photography is that of the object and subject: objects are the elements literally in the image, whereas subjects mean what the image is about; the themes you want to convey.
My idea: on Christmas Eve I was following Google’s Santa Tracker with a friend’s kid. Looking at all the ground Santa had to cover in 24 hours, I came up with the idea of Santa leading his troops to a near impossible battle against the forces of time and physics.
This is what we will dramatize. I want to see what types of setups and angles Midjourney comes up with, so we won’t be using a base image for this exercise. But if you have a fave Santa image, feel free to have a go.
3.1. Objects Define What Your Image is Of
Objects are the factual elements in the image; what the image is of; concrete things we can see. Every image has one or more objects, so we’ll start from these.
Key objects are the main focus of your image. This can be Santa Claus, a floating cloud or a herd of cats. Anything really that you want to portray.
In our example the key objects will be Santa Claus and some elves. There will also be a cathedral as a dramatic setting.
We’ll get to parameters later, and they always come after the prompt, but the one parameter we’ll introduce right away is ––v 4 which means we’ll be using the latest version of Midjourney.
Result:
Upon prompting, Midjourney will generate four variations for you to choose from. The numbers of variations run from top left to bottom right, row by row. You can use the ‘U’ buttons below the image to upscale any variant you really like. You can also create further variations from ones that are ok but not quite there yet with the ‘V’ buttons.
If nothing strikes your fancy, you can hit the ‘reroll’ button on the far right to generate a whole new set of images. Generally, if you don’t see anything you like on the first go, give it a few more spins before starting to tweak the prompt. This way you’ll give Midjourney a few more chances to express your prompt and get it right.
In our example, we wanted more variations from variation 3, but it didn’t ultimately pan out, so we upscaled variation 4 below.
Result:
Ok. So we got a Santa, a couple of elves and a sort of a cathedral there. But they don’t yet look like going into battle against the forces of time and physics do they? We need to start refining the subject of the image.
PRO TIP: to save time iterating prompts, I always copy the whole prompt before hitting enter:
The prompt may contain banned words and Midjourney will have you rewrite the prompt. You might not know these beforehand as Midjourney errs on the side of caution.
You might instantly after the first round notice things to fix or add in your prompt. This way you’ll have the original prompt ready to be pasted in a clean format right away.
3.2. Subjects Define What Your Image is About
Subjects are about themes, focus, atmosphere and other styling; what the image is actually about.
To start conveying the subject, we can introduce secondary objects, change their relationship to each other, and start introducing verbs and adjectives to draw out the topics and themes in a more refined way.
In our example we’ll introduce a battle axe, have Santa lead an army of elves, and make the cathedral vast to invoke a sense of grandeur. We’ll also try putting Santa and the elves inside the cathedral.
PRO TIP: when starting to draw out the subject, use very specific verbs and vocabulary instead of lots of general prepositions and vague expressions for the biggest punch in the fewest words.
E.g. ‘Santa Claus with a Battle Axe’ → ‘Santa Claus wielding a battle axe’ ‘In front of a crowd of elves’ → ‘Leading an army of elves’ ‘A cathedral with high ceilings’ → ‘A vast cathedral’
Better, but it all looks a bit static. I want Santa to be determined and on the move, as if the frame couldn’t hold him. I‘ll also add some armour, and inspired by Amazon’s upcoming Warhammer show, I’ll add that as a style.
PRO TIP: Again, be compact and direct instead of using a lot of filler words. Reduce redundancy and focus on pointed roles, intents and actions instead of vague ambivalent descriptions and prepositions.
E.g. ‘Santa Claus feeling determined and wearing an armor’ → ‘Determined armored Santa Claus’ ‘Posing in a dynamic manner’ → ‘In dynamic pose’ ‘In the style of Warhammer 40,000’ → ‘Warhammer 40,000 (style)’
We can add styling and adjectives anywhere, not just to the end; like we did adding ‘determined armored’ to the upfront. But keep in mind that things closer to the start of the prompt gain more importance. So save the beginning for your key objects and their descriptions.
Result:
Ok, we’re getting somewhere. We have a badass armoured Santa, leading an army of elves in a vast cathedral. And he’s clearly lit up and ready to go! But we’re not done yet. Let’s refine.
3.3. Further Refining Your Work
As we move on to further defining, detailing and stylizing our artwork, remember that you can add text anywhere, but reserve the upfront for the most important objects and subjects and leave things like general atmosphere, camera angles, and other style inputs towards the back.
In our example we’ll be adding the following:
Atmosphere and mood: what the setting and characters feel like in general from happy to trippy to brooding. For us this will ‘epic’.
Media type: from NatGeo photography to Renaissance oil paintings. For us this will be ‘highly detailed Unreal Engine 3D render’.
Result:
Looking great! But it’s getting a little heavy and losing the Christmas spirit! This is the Holidays after all, so we want to convey the theme of ‘christmas’ a little bit more.
Remember us talking about image URL weight in the very beginning? We can set custom weights to any text element in the prompt by using ‘::’ and adding desired weight after this. This is useful for e.g. setting an overarching theme we want to convey.
Remember that e.g. image URL weight is by default 0.25, so to overemphasize an element you can start from 0.5. The highest you want to go is 3 or so, lest the theme take over the whole image.
In our example we’ll add ‘Christmas’ in the beginning with the weight 1. Also, I’m noticing that as the different elements add up I’m starting to lose the elves in many iterations. We’ll add a weight of 2 to the elves too, just to make sure they stay onboard.
Result:
Ok! More ‘Christmas’, and we even got some candles in there. But I don’t think he’s really leading the elves now. They’re just kind of hanging around in the background. Let’s try and fix that.
Remember when I said things closer to the beginning of the prompt will gain more importance? To get the elves to follow Santa, we’re moving that part earlier and see if we can replace ‘dynamic pose’ with just ‘dynamic’.
Result:
Ok now Santa is leading the elves and we’re having everyone move to the same direction!
So there we have it. In a few iterations we went from basic objects to conveying the theme, a feeling and the atmosphere we want! Could we still improve on things, like ensure the axe is more refined and the cathedral more visible? Sure! But the theory should be fairly clear from here.
All that and we haven’t even talked about parameters, apart from ––v 4 which ensures we’re using Midjourney version 4.
Parameters are powerful tools for everything from setting canvas ratio to controlling more minute aspects of the work. They’re always preceded by ‘ ––’. They don’t need to be separated from the work with commas, and they can not have commas between them.
They always end your prompt, so once you start writing your first parameter with a ‘ ––’, then you’re done with your prompt and can only do parameters from there on out.
I’ll first go through the parameters I feel are most powerful and useful, and then cover the rest.
Version –– v (e.g. ‘ –– v 4’) This sets the Midjourney version you’re using. Default right now is 3, but many (like we just now) always use 4 as it’s more advanced.
Image Weight ––iw (e.g. ‘–– iw 1.5’) IW sets the weight with which the image prompt, if you’re using one, inspires the work. Default is 0.25, and 5 starts being very heavy, basically taking over the work.
Aspect Ratio ––ar (e.g. ‘–– ar 2:3’) AR sets the aspect ratio and is 1:1 by default. Note that v 4 can only do 1:1, 2:3 or 3:2, but these are plenty for most purposes.
Quality ––q (e.g. ‘–– q 0.5’) Q sets the quality of work and is 1 by default. 0.5 is half as detailed and costs less ‘fast hours’, while 2 is twice as detailed and costs more. 1 is usually more than enough for sketching.
Stylize ––s (e.g. ‘––s 900’) S sets a styling or flair value for the work. In v 4 it goes from 0 to 1000 and is 100 by default. It is not very predictable as ‘styling’ is very subjective, but test it out to see the effects.
Chaos ––c (e.g. ‘ ––c 90’) C sets how much the four variations created from a prompt vary from each other. In v 4 chaos spectrum is from 0 to 100, and naturally chaos 0 means little variation while 100 means a lot.
Negative Prompting ––no (e.g. ‘ ––no windows’) NO can be used to exclude elements from an image that are otherwise seeping in. E.g. you are prompting a ‘room’, and ‘windows’ seep in without you asking for them. ‘ ––no’ sets a negative weight of -0.5 to the word after it, and you can use this to get rid of these unwanted elements.
As an example, we might use the following prompt to get varying results:
Result:
Other parameters to explore (visit the manual for a closer look at parameters)
––stop Stop image generation at a given percentage before 100%, from 0 to 100%.
––uplight Upscale with ‘lighter’ algorithm that adds less detail when upscaling.
––seed Generate from a given seed, say when you want to keep variations close to original.
––sameseed Related to previous, generates all variations from the same seed.
––video Generates a progress video from image creation, which you’ll get in your inbox by reacting to an image with the letter emoji.
Settings You can invoke the settings menu writing the command ‘/settings’ in the input field instead of ‘/imagine’. This will pull up a menu where you can set default values for many but not all of the parameters above.
Preferences ‘/prefer’ allows you to set more options that are not included in settings, but that you’ll always want to use, meaning you’ll get to set your own render parameters and shortcuts.
References: https://medium.com/engines-of-value/get-creative-superpowers-with-midjourney-49ba5be174c6
