Mastering Midjourney in 2023 | The Ultimate Guide

mid-journey is complicated and it's ever changing it is a limitless tool for creating any imagery that you want but even on the latest version a lot of the time you'll get something that's just unimaginative and downright boring five months ago I made my most popular video on this channel it was called everything you need to know about mid-journey in 15 minutes or less I taught you about parameters and image prompts and weights and styles and so much more and together we tried to understand how to make beautiful art but as is the nature with the AI industry within a week there were several things about that video that were already outdated so to celebrate the new year I want to get you up to date and at the same time I want to turn you into an art generating Master to help you follow along I have a few introductions I want to make this is Roger everyone say hi Roger Roger wants you to subscribe he's a panda and he loves his safari hat you should also meet Hannah she's an elf and a princess and you may have noticed she also has purple hair lastly this is a mountain it's tall made of rock and has ice on top and we'll call it ice sea Peaks icy peaks we're going to learn the intricacies of mid-journey while generating these Concepts and trying to make them as good as we can let's get started the biggest difference I want to talk about is how there are now seven models being supported by the mid Journey bot you heard me seven let's type slash settings and look at these versions one two three four mid-journey test mid journey test photo and niji mode let's write a simple prompt for Roger and generate him in every version by the way for the rest of this video I'll be setting the seed to one for all my Generations that way when I make changes you can see exactly what those changes actually do the numbered versions work basically how you'd expect the higher the number the better the model is across just about every Dimension AI art has been improving a ton and version 4 is just newer so it's going to be just better overall mid-journey test is a model that I think is a bit of a mix between mid-journey version 3 and stable diffusion it came out in parallel with stable diffusion's open source release and it does a good job combining stable diffusion's knowledge with mid-journey's creativity my journey test photo is similar but it just seems to have been fine-tuned to make photos specifically the trade-off is that these test models are not nearly as intelligent as more classical mid-journey models and has trouble closely following the prompt notice that Roger here isn't even wearing a hat you'll also notice these test models generate two images rather than four for squares and if you make something that's not a square it only generates one but lastly we have niji and niji is a new mid-journey fine tune for anime and illustrative Styles and oh my God Roger looks great on to version four and it is the most popular model by far it is filling up that Community Feed and for good reason first off major and I recently made it the default even though it's still in Alpha but it is incredibly versatile the images look great and it takes simple prompts and turns them into beautiful pieces let's stick to version four and generate Hannah when we talk about the rest of the settings in this new UI by the way these settings are a friendly way to access the slash prefer suffix command and lets you automatically add parameters which we talked about in the last video to the end of all your prompts the first setting option we'll look at is quality the default value for quality would be Dash Dash q1 and it sets the GPU time for this generation to be about one minute if you lower the quality to half the generation takes half the time but also the details aren't as precise and Hannah looks kind of janky if you set it to high quality it takes twice the time but the details are generally going to be enhanced across the board for this image Hannah just looks better her background is more filled out and it might even be more clickable if I use her for a thumbnail okay on to Mid Journey's stylize function this wasn't here last time I talked about my journey in depth so here's an overview generally speaking low numbers will let your prompt speak for itself and mid-journey will not adjust the image based on its learned sense of beauty but higher numbers will do things like add makeup studio lighting or whatever it thinks will improve the image overall expect this to change and maybe even soon but at the moment the settings UI for style only really works for version three so let's show you how style works on all the models one by one well versions one and two don't even use it you'll get an error if you try version 3 uses stylized numbers from 0 to 60 000. let's quickly look at how these affect Hannah version 4 uses zero to one thousand the effect is a little less obvious here because it changes the most right at those very low numbers I'll let them finish playing so you get a sense of what it really does version 4 also has two style options versions four a and for B this is accessed in dash dash Style Space 4A or 4B this is because they weren't really happy with their initial implementation of stylized on this version but they didn't want to take the option away from anyone who liked it the test models test and test photo both use stylized numbers from zero to five thousand but I don't notice a huge difference on the flow take a look oh and niji mode doesn't support stylize at all either so moving it right along to the current options for upscalers the first thing to say is the default upscaler has been drastically improved you're way less likely to see newly generated artifacts than used to be but there are more added details that the upscaler can add and they're often photorealistic if that's the style of image you're working with the resolution from this upscale is actually smaller than it used to be it's 10 24 by 1024 for square upscales the light upscaler up light is essentially unchanged it's faster cheaper and not quite as detailed but maybe a little truer to the original generation if you're okay with a bit of a painterly feel though it actually gives more resolution for a square image it gives images that are 15 36 by 1536. lastly we've got the beta upscaler which generates the largest images it turns squares into an insanely high resolution of 2048 by 2048 which more than completely fills an HD 1080p display it's not quite as smart as the default upscaler is now but if you want a lot of pixels this is absolutely the way to go the last change I wanted to talk about is the improvements made to the concept of variations now when you select the setting remix mode you can change the prompt that you were using at the same time that you request A variation of an image this is a game changer it lets you be a lot more precise about the art variations you ask for if you generate icy peaks here you can get a variation of night time or sunset or Springtime or icy peaks in a lava World remix is incredibly useful as you follow a thread to make it something that is exactly what's in your head but beyond that I can release you from the lecture portion of this video now we're on to the fun stuff once you understand the settings and your options you have creative control to make whatever you want as you know prompts have three parts the image prompt the text prompt and the arguments if you want a master mid Journey one of those parts has a skill ceiling that will take you to the next level of course that is your text prompt how do you turn turn an idea into a prompt I'm going to teach you the concept of prompt engineering as I do it in mid-journey if you have a visual idea that you want to see come alive you can just type it into major names take a fresh lettuce leaf dripping with a gloop of oozing mustard you'll get some good results it's visual it's fun it looks like lettuce and maybe mustard but it doesn't have that wow factor and it just looks a little Bland I just want to remind you the seed is still set to one so if you want to try anything that I'm doing type slash prefer suffix Dash seed space one and you will get the same results that I do you can take that prompt a fresh lettuce leaf dripping with a gloop of oozing mustard and add some style Direction food photography suddenly it does look way better the lettuce pops and the mustard has a nice shine problem is I feel like I've lost that Oozy feel let's switch to a multi-prompt format and use weights so my style doesn't change the idea quite so much a fresh lettuce leaf dripping with a gloop of oozing mustard and then we put the multi-prompt breakpoint give it weight five and then put food photography and let's try weight three now that is Loopy but can we improve it at all the actual lettuce looks kind of odd let's add some tags for a little bit more fine control take a look fresh lettuce leaf dripping with a gloop of oozing mustard delicious food photography well-lit award-winning photograph appetizing composition let's toss in some specific camera settings and really solidify the image add some lighting info a bit of retouching and take a look at that top left Uzi goopy goodness my God it's beautiful this is why we make AI art that's for sure let's stop for a moment and think about what we just did this multi-prompt that we've slowly filled out actually has some cool properties namely the idea and the style are completely separate our lettuce leaf could be a painting or a logo it could be drawn in ink or it could become Vector art at the same time all of this stuff the style information it could be used for all sorts of food related images you could create an ice cream cone for celebrating Christmas or a cheeseburger with way way too much cheese though actually as we change the prompt you'll start to see some weird things are happening with the food we're photographing sometimes it's great but sometimes I'm getting this Smoky grainy strangeness that I would love to get rid of I've talked a little about this a little before but we can actually fix this in the multi-prompt by adding a prompt with a negative weight this can both reinforce the Style by putting a negative weight on ideas that clash and actually remove something you don't want in the image I can add those in and my ice cream cone looks way cleaner I'd say it makes this prompt as a suffix significantly more versatile let's look back at my final result for a fresh lettuce leaf dripping with a gloop of oozing mustard and just marvel at the glory of it now that I'm satisfied with all of these parts of the prompt I can actually save this prompt as a prefer option under the tag food that would be slash prefer option set then type food hit Tab and paste the whole string after the first prompt weight now if I have a new idea say a fruit salad with a lot of blackberries I can add that argument at the end that I just made dash dash food click enter and suddenly my whole multi-prompt actually fills itself out so yeah that was a bit of work but now quickly and easily whenever I have a food related idea I can use my food argument food parameter and get a beautiful enhanced result on top of that I already did this work for 25 different styles in case you want to copy off my test as I wrap this video up I will show you Hannah Roger or icy peaks for all of my other styles and you guys are welcome to use my prefer option list for anything you're making I really enjoy all of these that I created but I hope showing you how they're developed will inspire you to go step by step and make some styles that work for you and they should match your idea of beauty I think the most amazing thing about mid-journey is how it empowers everyone to discover the visual style that resonates most closely with them it might be quick and easy to type an artist's name and let their style control your Creations but that name is just a shortcut to describing the medium technique and skill that that artist uses if you want to create beautiful things in mid-journey I urge you to challenge yourself to find or even create a style that truly speaks to you mid journey is flexible and you have the power to control it just by typing words if you're finding that mid Journey's own style is limiting you setting that stylize option to zero will let you see your generation exactly how your prompt created it it can be humbling when you make something disappointing but it can also be freeing when your prompt creates something unlike anything you've ever seen I've created a paste bin with my output when I type slash prompt option list and it's in the description if you use any of them and want to share what you create while you're looking in the description I have a growing Discord community and we would love love to see what you made even more we would love to hear about a new style you stumbled across in your exploration good luck happy imagining and I'll see you in the next one [Music]

Channel: Glibatree

Views: 342,650

Keywords: AI Dungeon, GPT-3, AI Demo, AI Story Telling, Text Adventure, Gaming

Id: CBJFvrTAYi4

Length: 14min 15sec (855 seconds)

Published: Tue Jan 03 2023