Turn any idea into an image in seconds with Midjourney AI!

in a very short time generative AI has changed the way we perceive art so what if we want to combine our own imagination with this new technology how would we go about creating our own avatars but with our own ideas well in this video I'll show what I have learned about using mid Journey which is one of the most advanced publicly available generative AI algorithms to make your own avatars your own images application icons and a bunch of other tips and tricks well let's get started okay how the heck does it even work I'll try to explain how those algorithms work in the simplest terms I can considering the first sentence you usually hear when talking about diffusion algorithms is that it comes from non-equilibrium thermodynamics in physics it might not be that easy but I will try to make it as easy as I can if you don't care how the systems do what they do and just want to learn to control the magic feel free to skip to the next chapter most of today's genitive AI algorithms are done to a technique called gun or generative adversely networks it's a technique developed by Iron Goodfellow back in 2014 and has been under heavy researched since and lately it really opens a world of exciting opportunities in this family of algorithms you essentially pit two algorithms against each other a generator and a discriminator the discriminator is trained on billions of images to extract a mathematical representation from those images we call this an embedding basically the discriminator looks at an image and after seeing millions of cats the algorithms can say well yes this is a it I'm here live that's not I'm not a cat once I give the algorithms are pumped well tell me a cat the generator begins to generate a random noise for each part of the image and feed it to the discriminate until discriminator says well yes this is indeed a cat our image is this noise accepted by the discriminator since the discriminator fights the generator it's essentially its adversely that's how those algorithms actually got the name diffusion algorithms take this idea a step further well instead of generating just random noise they are trained on how the noise of billions of images look like and then apply specific algorithms that were attained on how to clean up noise from images so the whole results are a lot more intentional and look more natural essentially you could summarize these process in six steps step number one take billions of images on the internet with the text description of what's on the image the system learns how each image is encoded and how its corresponding label looks like in a mathematical representation called an embedding step number two take billions of images and slowly add noise to the images paste it lower under quality adding a little bit of noise every time around a thousand times step number three show the noisy images to an AI and teach it to clean the noise basically teach a machine learning algorithms how to remove the noise step by step to clean the image step number four generate noise that matches the mathematical representation of the image matching the pumps step number five clean up the noise getting a small but clean image and finally step number six upscale the image kind of like you know how the FBI shows this enhance in movies it's kind of like that scale up enhance this is by no means a comprehensive explanation and I will link in the description a number of videos and research papers that explain those Concepts in a lot more detail and in a more scientific way all right by now I think you're probably thinking enough with the Mambo jumper how the heck do I make an avatar with me Johnny well let's jump into that I went ahead and set up a copy of me Journey For You on the newly launched hiking Modern Life discourse so to tie all the examples from this video click the link in the description below enter the Discord server and go to the mid Journey channel so let's hide right now imagine pumped a boat sailing into that sunset by default me John uses version 3 but its new version 4 that is currently in beta is much more powerful and much more capable and we're going to be using it for all the examples in the rest of this video Let's tie the same prompt with version 4. to activate version 4 we just do hyphen hyphen V full in order to not have to add the minus minus V4 every time we can change it in settings plus MJ version 4 here and from now on it will add minus minus V4 for every pump we type you will notice that the journey created all those like eight little buttons below the images by default me joining generates four images and the four buttons on the bottom correspond to the images that generated U stands for upscale and v stands for variations if you press U on U1 it will upscale the first image here so one two three four let's upskill that one it looks the best if I pass V it will create variations minor relations on that image so V4 so those are the valuations on that last image and in a few seconds we'll see the upscaled version as well and there it is a beautiful image of a boat selling well actually from the sunset but will forgive the eye for that today now that we got the basics let's move on to some more advanced pumps first if you want to prevent something from appealing in the image we can always add hyphen hyphen no so first let's get rid of the cloud in the image imagine and we want to get rid of the clouds by adding the word photo we can make the images a lot more realistic because then the AI will use images that are actually photos that it was saying that foreign now one of the most powerful things you can do is experiment with the style the more detailed the style instructions are the more predictable the image becomes so let's tie a few examples now I want to remind you that the AI has been learning on a bunch of images from the internet and so if you add tags that people are likely to add to the photos you're more likely to get results similar to those images so for example photo of a boat sailing into the sunset filter realistic 8K 85 millimeters which is the focal length of the lens DSLR no clouds but the AI has also maintained on a bunch of classic artists and so we can use dull style as part of the image as well sketch of a boat sailing into the sunset Leonardo da Vinci style 8K all clouds also for example as another Da Vinci sketch of Superman because of course the AI has been trained on all kinds of Pop Culture images comic books it knows everything right we're talking here 5 billion images or for example homo Simpson in Van Gogh stalinite style right foreign night another very cool style is Unreal Engine if you want this 3D Lambert style kind of a realistic 3D landlord style let's say we do almost Simpson mango style night style and a little engine you can also control the light cinematic lighting or the Matic lighting will give you more of the light control make sure not to confuse lighting with lighting because that will look completely different take a look at what I mean foreign lighting and lightning not the same thing last part of this section each word you use can be given a weight by adding colon colon number so for example Flying Pig colon colon05 Cloud colon colon 2 will be a lot more clouds than flying pigs too clouds05 and it's very very different from flying pig no clouds by default everything gets the same weight and so the same importance in the generated image foreign pick 2 clouds 0.5 and you can see that the pig is the main subject of the image it's big it's in the center it's the main part this is fine Peak no clouds and this is how we essentially got flying pigs at an airport I assume or I don't know what the heck is that big flying but flying somewhere and you see that in this version that is flying pig these are five cloud two you basically don't see the Flying Pig anymore it's mostly clouds if you generate an image and notice that an image lacks a bit of small details you can use the Q parameter to control the quality if you give an image hyphen hyphen Q two it will work twice as hard on it keep in mind that walking fights and hard doesn't mean it will be much better since sometimes the I can overwork on an image and even though it's smaller detailed it actually looks less realistic or less beautiful to the eye but in some cases it can create really beautiful images and see here this is the Q2 version and this is the same pumped with q1 which is the default and you can see that especially here you see how much less detailed this image is compared to this with that said I still think the q1 in this particular case looks better you've probably noticed that all the images we generated so far had been perfectly squared that's a function of the way that the generative algorithm Works since it generates blocks of pixels however you can change the aspect ratio of the image with iPhone hyphen AR free colon 2 will generate a wide image two hyphen 3 will generate a tall image so let's now generate both of those and see the difference so this one is white look how realistically it rendered that bit here and this one is tall and some are modalistic and some are less realistic when an image is being generated a lot of the parameters that create an image all being generated randomly however there is no such thing as random in computers the all kinds of algorithms that come up with episodo random number that random number can influence a lot of the things in the image so what if you generated an image and you want to tweak just a small part of it me Johnny has a solution for that as well next to the image to the message in this code you have a little smiley face here press the envelope and within a few seconds mid Journey will send you a message with all the information about that specific image including the seed the seed is what actually is used to Generate random numbers but those images have been generated from so let's generate three images and try to use to use the C date so bad pit at a food market at dusk I don't know why I like bad pit as a subject today but why not so we're on this job and we learn it again and what will actually happen is that mid Journey will generate two random seeds and then I generate the images with those seats so those images are pretty different between themselves right now I go here and pass the envelope and now I got a message with that particular seed if I copy the seed and tell me Johnny to generate the same prompt and give it minus minus seed and the number we just got and what we got is essentially the same image now if I take that same seed and change the pump a little bit keep in mind that it won't be exactly the same image but it will be much closer and you can very clearly see here in this image similar oranges to baskets those two same baskets those oranges very very similar images except for the defense in the pump of course all right now let's get to the fun stuff mid Johnny can use other images as sauces or other sort of like inspiration the way to do this is by giving the URL for an image to meet Johnny and then adding a text pump as we'll learn above the easiest way to upload an image is basically to just drag and drop it here and submit keep in mind that you have to post the image in a separate message because otherwise you cannot copy the link so now you copy the link let's use the use the image we just uploaded and use part ownership at Sunset I want to be a poet sounds fun our pie is ready look at this part actually not a big fan of the hat but everything else is spot on right so let's upload a version of an image with white so you have one image with a white on the side and another image without the white on the side let's generate the same prompt with both images by default the image will have a weight of one and it's literally just installation me Johnny won't actually take and use parts of the image as base but rather take elements from it and incorporate them in the resulting image and here you can clearly see parts of the city part of the background and pretty a lot of white and here we see the port in the background since it's part of the image keep in mind that every little detail of the image will have an impact on the final result so for example notice that if the image has a bit of white on the side you can see that white in the bunker appealing as the city in the background and then let's take just the head and now we no longer have the city or the background well it's gone because it's no longer part of the original image still not perfect it still decided to add some windows but two out of four it did not because it did not get inspired from the original image and here's another cool trick if mid Johnny generated an image that is kind of close to what you want but you don't look like yourself in it you can feed me joining its own output as the source for the next image together with your own picture also when General avatars don't forget to add descriptors to the prompt so for example add beautiful or smiling or serious that will generate a very different image and here's the result is all that we get I don't know why I have long calculator if you look at the face here you'll see that the face is basically distorted all right now that you know how to generate awesome avatars you probably noticed that the biggest problem of the con generation of AI algorithms is the hands and the faces with hands there's currently not much you can do you can try to hide them you can tie to add in hyphen hyphen no on the hands you can try to explain for different pumps until something actually works or looks like human with faces however there's another gun coming to our rescue generative facial pile or gfb gun this algorithm essentially understands how a face looks like and when it notices any issues with the face it fixes them it's actually designed to restore old photos but it works really well on AI images as well if a face is a little bit distorted like the one here you just upload it to Jeff began with the URL for that Linked In the description as well it's C all you need is a GitHub account and you submit it and within a few seconds you get the face fixed that looks much better than the aigelated one I realized that might be a lot but still that's not nearly everything that me Johnny can do for example you can use me Johnny to generate an icon for your app this is our application icon I mean this is crazy especially after upscaling a bit look at this icon it's really beautiful you can generate beautiful dashboards or like use interface design ideas with a simple prompt as well and there is a beautiful dashboards I mean can you imagine using an app that looks like this that is a completely beautiful or you can ask me Jonah to generate a beautiful landing page make sure that you use hyphen hyphen ar322 in order for it to be landscape instead of a Squall and this is all beautiful landing page of course all of those are just inspiration okay since you watched all the way until now here's a bonus tip what if you already have an image that is almost what you want but you don't know the pump of it well how about a quick way to recreate a pump from an image meet pound to IMG can just upload an image submit and within a few seconds we get a prompt that can generate specifically that image or something very close to it and this is what we got a foggy park with benches and these tilt shift photo by Mike pilk Shutterstock contest winner tonalism soft Mist High dynamic range and mist and apparently that should be enough to generate that image in mid Journey all right that's been a lot I know but I really hope that you learned something from this video genitive AI is here to stay and it definitely came with a splash be sure to visit our Discord community and share the most amazing things you can do with this and other algorithms and to play for yourself with all those amazing pumps speaking of algorithms well don't forget to like this video to tell the YouTube algorithm that more people should see it and subscribe to this channel so you won't miss any of the future digital life hacks and until next time see ya

Channel: Hacking Modern Life

Views: 310,078

Keywords: ai art generator, artificial intelligence, ai, artificial intelligence art, midjourney, ai generated art, ai art generator app, discord ai art, ai art generator website, ai art prompts, dalle 2, guide to midjourney, text2image, text-to-image, lensa, lensa ai, openai, midjourney tips, how does ai learn, discord midjourney bot, how to make ai generated art, discord ai bot, artificial intelligent art, jasper art, ai image, midjourney ai, ai art generator free, midjourney ai tips

Id: znzt_3fZdYg

Length: 19min 58sec (1198 seconds)

Published: Fri Dec 23 2022