I know it can feel overwhelming when
trying to get started with using the Midjourney Website. There's so many
different buttons and options, it's hard to know what you're supposed to do. Don't
worry, let's walkthrough how to use the Midjourney website step by step. This
will be a simple and easy to understand guide, and I promise that by the end of
this tutorial you'll be able to use all the features of the new website. Hey Tao,
do you prefer working on the Midjourney website or Discord? Definitely the
website. It's much easier to navigate and organize your images. The interface is
also very intuitive, so I'd recommend anyone who has access to the website to
give it a try. Once you've set up your account and logged into mid journey.com,
this is what the homepage will look like. We're in the Explore tab right now, where
we can see images that other creators have generated. On the left there's a
couple different tabs available to us. The Explore tab we're currently viewing,
the Create tab where you'll be working in to generate images, and the Archive
tab to organize and sort your images. Then there's a couple more pages to
rate images and give feedback to the Dev team. For now, let's dive right into
generating images. But first, i'm going to switch the color theme of the website
from light mode to dark mode with this button in the bottom left. This is just a
personal preference, but I find it much easier on my eyes. To start generating
images, let's go to the Create tab that looks like a paintbrush icon. The
Create tab is your main workspace for Midjourney. Here's where you'll prompt
for your pictures and customize them to your liking. The images generated for
each prompt are neatly organized into rows with the prompts and parameters on
the right side of the page, and the corresponding images in the grid on the
left. You can scroll up and down to look through previous images you've generated.
To start creating images, use the Imagine prompt bar at the top of the page. Since
it's close to summertime where I'm at, I'll enter the prompt photo of "Lake
Tahoe on a clear summer day, there's a man on a canoe." Once I enter the prompt,
I'll see the progress of images being generated below. And like I mentioned
before the prompt with our summer lake will be on the right of the page. Once
the photos are finished we can immediately create variations of them by
hovering over the images. The Vary Subtle button generates new images that have
the same structure and color of this photo, but with some slight changes in
small details. While the Vary Strong button creates variations of this photo
with more obvious changes in composition and larger details. If we take a look at
the photos generated with Vary Subtle, they more or less look the same with
minor differences in the location of the canoe. But the photos created with Vary
Strong have much larger changes in the geography of the trees and mountains in
the background. Now that we've created our first images, let's take a look
at the Settings menu which gives us all kinds of different customization options
for our images. I'll prompt for "a 3D render of an isometric Asian Shrine.
Created in Blender." To access the Settings menu, navigate to this icon with
three horizontal bars and circles inside the image tab. The first thing we can
change is the Image Size or Aspect Ratio. You can switch between a vertical
portrait image, the default square image, or a horizontal landscape. You can also
drag this slider underneath to get even more options for the Aspect Ratio. I've
generated some images with a default square 1:1 aspect ratio, then a wide 16:9 aspect ratio, and also a vertical 2:3 portrait aspect ratio. Notice how
underneath the prompt is this small bubble that shows the Aspect Ratio
settings we used. All the custom settings will be displayed here. Click on it to
use that setting inside the prompt box, and --ar 2:3 is pasted into the prompt.
The two dashes (--) followed by parameter name and value is a shortcut way of
using custom parameters in Midjourney without the settings menu. In this case
it generates images with a vertical 2:3 aspect ratio. Let's take a look at
some other settings in the Aesthetics box. We have Stylization, Weirdness and
Variety. Hover over each option to see more information about them. Stylize
controls how much creativity and imagination Midjourney injects into the
images. A lower Stylize value matches a prompt better but won't be as
artistic. Increasing Weirdness creates quirky and unconventional photos. These
images are probably not going to match what you prompted for but you get some
interesting results. Finally there is Variety which used to be the Chaos
parameter, and controls how how much variation is in the images being
generated. Let's take a quick look at some of these different parameters at
work. On the top is our 3D Shrine generated with the maximum stylize
value of 1000 and on the bottom the stylized value is set to 0. The top
row with high stylize values has way more intricate architecture details with
trees, bridges, and water, although it may not look as similar to the game engine
asset as the bottom row. The bottom row has way less detail and
the results are fairly basic. Here's a comparison between a high weird value of
1000 on top and a low weird value of 0 on the bottom. In this case it looks
like a high weird value produces many different architecture styles. We also
got some buildings that look like they were constructed from wood. I kind of
like this tall pagota looking building on the left. And finally here's a look at
the effect of increasing the variety or chaos of the image.
I don't use this parameter as much because it reduces the level of fine
grain control you have from your prompts. The photos generated with a high chaos
value hardly look anything like the shrine that we asked
for. So I've entered a new prompt for a coffee shop generated with anime style
art. If you hover over the parameter in the Settings menu, on the right side of
the info box it tells you the parameter shortcut you can directly enter into the
prompt. For Stylization this is --stylize. So let's try adding "--stylize 300" into
the prompt. If we hover over weirdness we see that the shortcut is --weird so
let's also add in "--weird 20" into our prompt. Now next to our images we can see
all the different parameters we've used, of "aspect ratio", "stylize", and "weird". I'm
also using "niji 6" which is a specialized Midjourney model for anime and
illustration style images. We can switch up the Midjourney model that we're
using in the model tab. The two I'd recommend you use would be version 6
which is the newest default Midjourney model and produces the highest quality
images. And Niji 6 which is the anime style model. The Niji 6 model is really powerful
and I've had a lot of fun prompting for anime style images with it. The mode of generation you use is
especially important when trying to create specific art styles. Raw mode
offers a more direct interpretation of your prompt, without injecting as much of
Midjourney's automatic image enhancement into your pictures. For
example if we prompt for Studio Ghibli style images and we use Raw mode, which
is shown in the top row, it produces images that aren't as bright, saturated,
or colorful compared to the Standard Midjourney mode shown in the bottom row. But
they're much more faithful to the original Studio Ghibli style.
You can also control the speed of image generation. This depends a bit on
your subscription plan. I just stick with Fast mode which works fine for me. But
you can use Relax mode which is slower but doesn't consume GPU hours, or use
Turbo mode to create images faster but it'll burn through your GPU hours faster.
The GPU hours are basically like credits. There's a couple of other hidden
parameters you should know about. First up is Negative Prompting, which lets you
specify what you do not want to be in the image. Take these landscapes
generated with the prompt of "a serene landscape in a fantasy setting. In the
style of Studio Ghibli." The images are very pretty, but what if we don't want
any mountains in the landscape? In that case we can use Negative Prompting.
Inside the prompt box add the parameter "--no" and after that put in all the
things that we don't want to appear, so in this case I'm going to add "mountains."
Then when I generate the landscapes Midjourney will avoid generating photos
with mountains. Negative Prompting is a super useful parameter that you should
definitely be taking advantage of. Next up is Tile, which lets you generate
repeating patterns for fabrics or wallpaper. Just add "--tile" at the
end of your prompts and Midjourney produces a repeating pattern for you.
Then to create the full tiled pattern, open up an image and download it with the
downwards arrow button. Don't worry, we'll cover all the other options here in a
second. Then head over to pycheung.com/checker. I'll put the link in the
description. Upload the image we just created and then drag the slider around
to change the size of the pattern. When you're happy with it you can download it
for free. This is one of the most underrated features of Midjourney. I'm
always surprised to see how nicely the tiled patterns turn out.
The Permutations feature will save you a lot of time when you're prompting,
and it lets you iterate through your designs very quickly. It's accessed
with these curly brackets "{...}" in the prompt shown here. Inside the brackets you can
separate keywords or phrases with commas. For instance the different colors of
{blue, orange, green, red} and Midjourney will run multiple versions of
the same prompt with the different colors at the same time. So we don't have
to run four different commands separately. We can also use permutations
for parameters like changing up the amount of stylization that's
applied. In these images of the female astronaut, you can see that it slowly
becomes more and more stylized. On every page you have the
search function in the top right. This is the quickest way to sort through your
images. If I search for a word like "dolphin" it brings up all the images that
I've generated with that keyword in my prompt. In addition to normal keywords,
you can also search for prompt parameters like "--ar 9:16", which pulls
up every image you've ever generated with that aspect ratio. There's also a
small menu underneath the prompt on the right side. If you hover your mouse over
this area underneath the prompt, a couple icons pop up. The Rerun option with the
circular arrow will rerun the same exact prompt and settings. This is always worth
trying if the images don't look exactly like what you want the first time around.
It can take a couple of attempts to get a good image to work with. The use option
which is a capital T icon copies and pastes the prompt, settings, and any
attachment into the prompt box on the top of the screen. It's a convenient way
to make any adjustments to the prompt without having to type it all in again.
You can also hide images. This doesn't delete them and you'll be able to find
them again later on, it's just a easy way to clear up your workspace.
There's a couple other options here. You can download the full image
grid with all four images or open these images up in Discord if you want to work
on them there. You can also copy the prompt and image from here. Let's not
worry about all these options for now. There's a lot that can be done with just
prompting alone. If you want to see a more in depth tutorial about prompting in
Midjourney, I'll link a tutorial video guide that I've created in the
description that covers everything from different colors, textures, subjects,
materials and much more. Moving along, let's take a look at all the
different ways we can manipulate a base image once we've prompted for it. Click
on an image in the image grid to open up the features menu. On the left you'll see
the base image and on the right is all the different ways we can manipulate and
customize it. We've already taken a look at the Vary Subtle and Strong buttons
which generate some variations of our image. The Upscale options lets us
increase the resolution 2x. So a regular square image with a resolution
of 1024 x 1024 will be upscaled to 2048 x 2048. Upscale Subtle enhances the
resolution but remains faithful to the original image. It keeps the original
details as much as possible. Upscale Creative also enhances the resolution by
the same amount, but it makes minor adjustments to the fine grain details of
the image in an attempt to improve the image quality. If you like the base image
already, then you probably just want to use Upscale Subtle to not introduce
random new details. But Upscale Creative is really useful for images like this
one. I've generated a photo with multiple people in it
and if you look closely the faces don't look quite right. The eyes clearly show
that this is not a real photo and is Ai generated. They just look weird. Here's
what it looks like when the Upscale Subtle feature is used. The resolution is
enhanced but the facial features haven't changed much and the eyes still look wrong.
If we check the Upscale Creative image though, the faces look way better. The
eyes now look more natural and human. It's not perfect and there's still some
slight issues, but it's way better than the original image. By the way next to
your prompt on the right side there's these red text bars that tell you what
types of features you've applied to your images. Whether that's an Upscale or a
Variation or anything else. The Remix feature does exactly what it sounds like.
You can take the general composition and subjects inside this photo of hikers and
evolve it to have a different setting or style. Let's see an example. I'll use a
Remix Subtle feature which is going to paste the prompt into the prompt bar, and then
you'll see the image attached with the red text Remix to let you know you're
using remix mode. Now let's add in that their in snow covered mountains and
wearing winter hiking gear. This should generate an image that keeps a similar
general composition but switches up the environment. You can definitely see the
influence from the original photo in the remixed images in terms of the subject
placement. There's also the same angled slope for the mountain in the background,
although now it's covered with snow. If we use Remix Strong, it'll generate more
drastic changes between the newly generated images and the original images,
but you can still see how the people inside the photos look quite similar to
the base image. Next up are the Pan and Zoom
features which let you expand and change the aspect ratio of your image in
different directions. The Pan feature extends the image in directions
corresponding to where the arrows are pointing. Using the up and down arrows
extends image vertically and using the left and right arrows extends the images
horizontally. It changes the aspect ratio and expands the canvas without changing
the content of the original base image. Then there's the Zoom button which will
pull the camera back and expand the image in all directions. The expanded
canvas is going to get filled in using context from your base image and prompt.
This might lead to some pretty unexpected effects though, like the
floating architecture in some of these images. The Change AR feature is like a
much more customizable version of the Pan feature. It lets you directly choose
the aspect ratio you want to expand to by using the slider at the bottom. The
gray outline box around the base image shows how the borders of where the canvas is
going to get expanded to. Then we can choose where we want the original base
image to be placed inside the new expanded images with the Start, Center, and End
options. This is kind of like the Custom Zoom button in Discord but with more
control over the placement of the original image. The best part about this
is that you can use the Pan and Zoom features as many times as you want. As
far as I can tell, you can't combine the Pan and Zoom features with the Remix
Mode on the website like you can in the Discord Midjourney version, but I'm sure
they'll add that soon. We've already discussed the Rerun button which
just runs the same exact prompt again. The Vary Region button is inpainting. It
lets you edit specific parts of an image. You can use either the rectangular
selection tool to select a grid or use a lasso tool to select a more refined
region. If you make a mistake in your selection, just right click to undo. Then
inside the prompt box below, we can edit the prompt adding whatever new objects
or details we'd like. I'll add in an ancient pyramid structure in the
background that's covered with plantlife, moss, and flowers with a large
stairway. Once we enter that prompt, we'll get confirmation that the generation job
is started. And here's the new images with
the Inpainted pyramid. The Vary Region tool is especially
useful for photos that Midjourney can't naturally generate. What do I mean by
that? Here I prompted for a "fullbody shot of the Joker wearing Spiderman's
costume." I wanted him to be wearing the iconic Spider-Man outfit, but for the
most part it just generates the Joker wearing a jacket and pants with red
colors and stripes. But the Spider-Man costume doesn't really show up in these.
Instead we can try generating a photo of Spider-Man in his costume and then use
Vary Region to inpaint the Joker's head onto that. With this photo of
Spider-Man, I'll use the Vary Region tool to carefully select the shape of where I
want the Joker's head to be. And then change the prompt underneath to "the
Joker". A couple of the resulting images weren't able to successfully add the
Joker to them, but the first and last photos worked okay. And we see the Joker
in a Spider-Man costume, up to no good I'm sure. In the last row there are
features labeled "Use". We have Image, Style, and Prompt. Image uses this base image
as an image prompt. Click on it and inside the prompt box you'll see the
current image as an attachment. Style transfers the style from the current image
into a new image, which lets you generate multiple images with a consistent style.
And Prompt simply copies and paste the current prompt into the prompt box. We'll
cover how to use image references in the next section of this tutorial. For now
let's finish up with the other features here. This downwards pointing arrow is the
Download icon. The magnifying glass is Search and it finds other images from
other users that are similar to the current image. It looks like there's a
lot of other people people making photos of the Joker, although we're the only
ones giving him a Spider-Man costume. Finally you can Favorite the image which
makes it easier to find later on. By the way, if you Right Click on any image in
the Create tab, it'll show you a menu with most of the image manipulation
features we've talked about. You can't use Pan or Zoom here but other features
like Vary region or Upscaling can be done. If you're just getting started with
Midjourney or need some inspiration for what types of prompts to use, go and
check out my eBook guides that cover everything you need to know about
prompting. Some of these are completely Free and Easy to understand, so make sure
to pick one up. I'll post the links in the description. Now, let's move on to
using images as references. Midjourney has come a long way from when it started
where there weren't really that many options for image prompting. But nowadays
you can use images for generating Consistent Characters or getting a
Consistent Style for all of your photos. This section of the tutorial will cover
all the different ways you can use image references in Midjourney. To upload a
reference image use the plus "+" icon in the prompt bar. From there we can choose a
file from our computer to upload. I'll start with this image with a silhouette
of a man and light rings behind him. The first thing we can do is ask Midjourney to describe the image for us. Just hover over the image, and click on
this icon with a letter "i" in it. This generates a set of phrases that describe
what the prompt for an image like this will look like. We can pick and choose
from these and add them into the prompt box. It won't generate the same exact
image for us, but it'll give us some ideas for how to generate similar images.
This is a really useful function if there's a specific image you like, but
you're not sure how to prompt for a similar image as that. Another way to
upload reference images is to just directly drag and drop them from
the website. If we upload multiple reference images, we can actually Blend
them together. So I'll drag in this picture of a shrine and then next to it
I'm going to drag and drop this image of a landscape photo of a cherry tree
floating in the ocean. Since both images have a portrait aspect ratio, I'll add in
"--ar 2:3" so that the Blended image will also have a portrait orientation. When we
go ahead and generate the Blended photo, the result will be some combination of
the shrine and the tree in the ocean. You can blend more than two images. I don't
think there's actually a limit anymore, but the more images you blend the less
the final resulting picture will look like any of them. We can use image
prompts to embed all sorts of different concepts into our images such as
textures and patterns. I'll upload this image of a
geometric pattern and then prompt for "fullbody shot of a runway model wearing
a dress. She has brunette hair and is wearing heels --ar 9:16". The results
we get do look inspired by the reference photo we've uploaded, but we can't see
the head of the model and it's hard to separate the floor from the model's
dress. To deal with this, we can adjust the Image Weight which
controls how much influence the reference image has on the generated
photo. It's used by adding "--iw" to the prompt and for version 6 this is a
number between 0 to 3, with the default being 1. Let's try to turn this
down a bit. I'll use the capital T icon to paste the prompt and reference image
into the prompt bar, and then add "--iw 0.3" at the end. Now the generated images
won't have as much influence from the reference image so that the figure of
the model is much more clear. But we can still see a lot of inspiration from the
image prompt in the colors and geometric patterns on the dress.
We can also use image references to generate multiple photos with the
same Consistent Style. Here's a pretty well known art piece. After attaching it to
the prompt box, if we hover over the small picture icon, three options pop up.
The first option which looks like a humanoid icon is the Character Consistency
feature. The second option that looks like a paperclip is for Style Consistency,
and the third option which looks like a picture is to use it as a
normal image prompt. Let's select the paperclip icon to use it to generate a
Consistent Style. Now inside the prompt I'll use "retro wall art of a spaceship"
with a 2:3 aspect ratio. When I run this job, it should
generate posters of spaceships but with the style of the reference image. And
that's exactly what we get. You can clearly see the influence from our
reference photo in the generated images. I repeated this process for a couple of
different prompt subjects like seahorses or cactus plants. And there's that same
consistent vintage asian style for all of them. A cool thing about using image
prompting is that you can upload multiple reference images for Style
Consistency. Here's a set of images I've collected that represent a Rustic
Countryside Vibe. I'm going to use these bright cozy images as style references
and generate a variety of different photoshoots to see what Midjourney comes
up with. Let's drag and drop them into the prompt bar. Then select the paperclip
icon for each of them to use them as Style References. Then we'll add a prompt
of "Photo of a two-story house. There's a cozy and rustic vibe". This generates a
set of photos inspired by our five style references and we do get a pretty cozy
looking cottage with old architecture. Using those same exact
style references we can prompt for the other rooms in the house like the bathroom.
I like how the third image here has some small light decorations.
And here's a kitchen. We've also got a home office to make YouTube videos.
And finally we have a cozy living room. You can build an entire world using
these same style references. One of the most popular
recent features is the Character Consistency feature which lets you
create photos of the exact same person in different environments, with different
facial expressions, and from different camera angles. I've generated some base
images for this young Hispanic woman wearing a pink sweater. To use one as a
character reference, drag it into the prompt bar and then we'll pick the
person icon for Consistent Character photos. I started by pasting in the prompt.
Let's try using a different camera angle like a "side profile shot
from the side." Let's give her a happy expression and we'll also put her in a
new environment like a garden outside during golden hour for nice lighting.
The generated images won't always resemble the reference photo you use.
The Character Consistency feature identifies key facial features in the
reference photos and generates that, but sometimes it can produce a caricature
type of effect. So you might need to try a couple of different reference photos
to find one that works well. A key parameter you need to know about is
Character Weight. It's a value between 0-100. It controls how much creativity
you can inject into your images. The default value is 100, which focuses on
replicating the face, hair, and clothing of the reference. But the smaller you
make this value, the more freedom it gives you in controlling the clothing,
hairstyle, and actions of the character. Why don't we prompt for her to
be reading a book in a library, but this time wearing a white collared shirt
instead of her pink sweater. Then I'll add in "--cw 10" to turn the
character weight down. This allows us to change her attire and in the resulting
images she is wearing a white collared shirt just like we asked for. If we take
a look at what happens if we use a default Character Weight of 100 you can
see that Midjourney isn't able to switch out her clothing with the white
collared shirt. One of the most common questions I
get asked is if it's possible to generate multiple consistent characters
in the same photo in Midjourney? I've posted a video tutorial for that, so I'll
also link that in the description if you want to take a look. The best thing about
the Midjourney website is how easy it is to organize and sort through all of
your images. The Archive tab is where you can easily find specific images or put
them into folders. The first thing you want to customize is the view options.
You can leave them as a default Square thumbnails layout which crops your
images to squares, but I personally like to change it to the full layout so I can
see the entire image. We can also adjust image size displayed on the screen. Next
up is the Filters tab where you can choose which types of images you want to show.
For example, I may only want to see images that I have Upscaled and have a
vertical portrait orientation. You can always also use the Search tab to narrow
down the images displayed even more. I might only want to see upscaled portrait
images that have the word "tree" in their prompts. The Folders tab is a great way
to organize your work. Let's create a folder and give it the name "ancient
ruins." We can then directly drag and drop images from the viewing gallery into
that folder. So let's add in a couple to test. And then if we open up "ancient
ruins" we'll see those images inside. We can also make smart folders. That lets us
organize images by prompt keywords. I'll create a new folder titled "instant
cameras." Then if I check the smart folder option. I'll be able to enter Search
terms. If I add in some keywords here like Polaroid, Instax, Disposable camera
all separated by commas, it automatically adds all photos created with those
keywords in the prompts to my instant photography folder. It'll take a second
and you might need to refresh the page. This is a really nice feature that makes
organizing a whole lot easier. Let's head back to the Explore page where we
started this tutorial, to see what types of images that other people have
been making. You can take a glance at which photos have been popular recently.
Any picture here can be used as an image reference and the prompts and parameters
can be pasted into your prompt bar. The magnifying glass finds other similar
images that people have been working on. We can also visit someone else's
profile page and take a look at the types of images they've been
making. If you're looking for inspiration use the search bar to find examples of
what's possible with the newest version of Midourney. Here's what comes up when
I search for "watercolor wall decor." You can also give direct feedback
to the dev team. One way is to rate images to help improve their datasets
and model training. You can get some free GPU hours if you're one of the top raters.
In the page underneath that you can also submit ideas and see what the
most popular feature requests are. These will probably be the ones that they're
going to implement soon. And also you can check for the
latest updates. Here's a couple of frequently
asked questions. Can I use my own images for the Inpainting pan and zoom
features? Not right now, there's not really any way to edit your own photos
right now, probably because of legal and privacy concerns. Can I use my own
illustrations for the Consistent Characters feature? You can try uploading
your own photos for the Consistent Characters feature but it's probably not
going to work that well. Midjourney is only designed to use that feature with
images generated by Midourney. So now that you're familiar with the Midjourney website, go and take a look at this guide on prompting for "Cinematic
Photography in Midjourney" if you want some tips on how to prompt for the best
looking photos. Also make sure to Subscribe if you want to see more like
this.
