Unbelievable Realism with Midjourney AI v5

mid Johnny has positioned itself as the leader in the AI generated Art Space and even though the space became very crowded between stable diffusion Dali Blue Willow and other players that just recently entered the space my journey is still leading the pack and now after months of testing and tile me Johnny have introduced its version 5. so what's new well let's find out [Music] a few weeks ago I published my tutorial on meet Johnny 4 and it's been tremendously successful it had over 200 000 views and bought over 3000 people into the hiking Modern Life Discord by the way join us in the Discord as well in this tutorial I taught how to use my journey from beginner to an expert including how to generate your own AI Avatar and a lot a lot more be sure to check out this video we'll have a whole detailed tutorial on me joining Channel me Johnny has at least four versions of 2022 version 1 and July 2 in September 3 in August and for November and version 5 has been released in March 2023. for the past few weeks the mid Journey have been publishing all kind of images from mid-jorn E5 and letting people lead them to see which ones are the best and now finally me Johnny have allowed us all to use that new version so what is new first you will notice that hands are mostly fixed for example this photo of a woman or the evening goodbye notice that most of them I think actually all of them actually have five fingers and a purple structure of the hand let's upscale number four heal and you see a purple hand compare that to those hands heal which are basically nightmare fuels for example that number three or that number one well the hand is flipped all the fingers are cut and so on generally I is notorious for how hard it is to generate fingers and version 5 is much much better than anything before however it's still not perfect majority five images also have a way wider dynamic range which means the difference between the dark areas and the black team areas of the image so for example check this one out let's start with version four and then we'll switch to version 5. you can see here that the color and the images is rather flat right the background and the shadow are pretty similar in the brightness compared to version 5 where you can see that the blacks are almost exactly blacks and the white became very very bright pretty beautiful I just have some chills down my spine just looking at that image B25 is also much better at understanding natural language so so for example if I say a cat eating in ice cream I really don't know what's up with me in ice cream today I guess I just really want ice cream I kept eating ice cream it's very different than cat and ice cream or cat and ice cream check that one out that one is super cute okay here we have a cat with a double deck of ice cream it's crazy how realistic everything is except for that idea of the second bowl of ice cream is kind of stuck on top of the first one then that one came out as a like the cat bean and ice cream and here we have the cat and the ice cream basically like looking at each other very very different me Johnny is also much much better at generating a realism and in fact it will go towards more realistic images by default interestingly enough with that emphasis on realism mid Johnny 5 also can emulate a lot of different lighting conditions and lenses with very different lighting very easily to get full control over me Johnny 5 it really helps to have all the different photography terms nailed down like for example what is the difference between backlight and font light flat light versus remember in life and also the different lenses that could be used in order to make all of that easier I will link in the description my light and lens guide that will give you examples of generated images with different lighting and different lenses that you can use in your own images for example a dark little medieval tongue with volumetric lighting V5 versus Spotlight and now let's add lens instructions so and something closer and let's compile all of this and how it looks and how it behaves so here you have a medieval time town with volumetric light now we have some Spotlight and you can see it does a lot less of this fog and kind of diffused light but other more precise and at 25 millimeters you see a wider kind of image from closer it almost looks like toy houses compared to that that looks a lot more realistic on maybe like a 3D video game until you have a 85 that kind of zooms a lot more and again with spotlighting it's finally enough that me Johnny decided to make it more of like a puppet Town even though we did not give those instructions in the pump itself the Jolly 5 is also a lot a lot better at complex plants so for example anthropomorphic cat Pixar style similar in style to the Last Supper sitting on the table well in case you don't know what antrimorphic means it means human-like playing poker in a dark smoke your room inspired by realistic animation elements from pants label it and not only Guillermo del toy Works scary and cute at the same time let's hunt this and see what we get and you can see I use the aspect ratio of 16x9 so a white skin aspect to Asia and it becomes very stylized and very very realistic and actually what I imagined does our anthropomorphic cats Pixar style playing pocket in the style of the last supper or for example knolling knowling is always finite nulling is the art of putting objects in a pattern so let's see how V5 does it and here we have beautiful nodding of photography equipment thank you Jean for sharing with me the nulling pumped that is really awesome except that lens is pretty crazy and look it almost got Canon properly written here another thing that was fixed in a mid 25 compared to me Johnny 4 is and the aspect ratio and now mid Johnny 5 basically supports a huge variety of aspectation it's no longer limited to two to three or three to two like mid Johnny 4 has been and you can even do on a morphing aspect ratios which basically means like the cinema usually it's 249 by one however me Johnny does not support floating points or like decimal point inside aspectation and so you can just multiply it by 10 so basically instead of 239 by one you write 24 by 10. and here we have a Blue Smurf plotting Revenge at a anamorphic aspectation mid 25 is also bringing back the tile parameter that allows you to generate tiling images which is pretty cool if you want to do textures or repeating wallpapers or any images that kind of repeat on the corners or on the edges check this out so let's generate an audio active detailed class with mushrooms and make it tile and you can see that those images leave Pete on the corner you see the piece of mushroom heal here is the continuation of the mushroom heel and heel and heel is the stem of that mushroom one of the most popular uses of me joining is to do AI self portraits and me Johnny 5 have become much much better at that it's battle replicating images and it's a lot more flexible and because of its wider Styles and dynamic range it can generate much better images and in fact later in the video I will show you how my journey itself can extract pumps from existing images and use them as well I promise you that they will show how to upload images and copy the URL on mobile on desktop and inside the Discord app so depending on where do you have this code open the process is slightly different but not too different so if you're inside the Discord app it's actually pretty simple you can either place heal on the plus upload a file choose an image send it and then right click and copy link and we have the link to the image make sure that you have attachments in the resulting image otherwise it's not the right URL if you're running this code in the browser so you see the URL here that's your comb or Edge or whatever the process is very similar you press on that you choose the file upload it however you don't have here the option to copy the image so how do you do that you click on the image and you get this pop-up with the image inside click here copy image address and then you go back to here and you have the same attachment URL if you own a mobile plus plus here choose the image and upload then long pass on the image itself and copy media link and then when you passed you see the same attachment here now let's tie the same image we did with version 4 before so that URL has a pilot read version 4. and you see now that me Johnny shows the image that you just copied and let's compile it with version 5. and here you can see that actually the image is less similar to me than in version 4. mid Johnny 5 also finally supports image weights again what are image weights image weights allow you to Define how much weight the image has as opposed to the text pumped kind of similar to how I showed earlier the double semicolon 2 or double semicolon 1 in text pumps the same with images so how do you use it well you add minus minus IW between 0.5 to 2 and the higher the number the closer the resulting the generated image will be to the input image although that as a medium color to the image pumped so if you're using your own photo for a portrait or like for an AI Avatar and it doesn't come out similar to you at all well type to use an image weight Which is higher if you want something more creative and kind of more animated or cartoonish use an image weight which is lower let's take let's take the image free here and try to make it a bit more similar let's make it in a jungle that one seems to be the most similar to me let's upscale it and I want to experiment a bit with image weights so Image Weight 0.5 versus Image Weight 2. and here you can see very very clearly that with 0.5 the images don't look like me at all in fact this one looks like a girl with long hair for some reason but at the two the images are much much closer to me however the face does not look correct and we will fix that later what if I want to make comic book well the mid Johnny 5 realistic training actually works was fast here let's compare version 4 and 5 at a comic book illustration in a jungle and you can see that version 4 really looks like a comic book illustration basically a comic Google station of me version 5 however because of its most modelistic training looks a lot more well realistic it kind of looks like a comic book but then it doesn't the background looks more realistic and I look like a figure rendering version of myself alright I promise you a few bonus tips so bonus tip number one is slash describe slash describe allows you to extract the Pumped or like on a helmet Journey sees your own image pumps and then choose pumps to generate from this image and essentially you can even modify change it or combine different terms together so let's say that same image we generated earlier admit journey is looking at that image and trying to describe what it is with its own prompts and now we can choose a specific pump Hill and generate images from them so in this case we see the man standing near river in the style of social media portrait Kia and Gill Allen I have no idea who does that means the atmosphere Walt Disney webcam photography historically accurate hyper realistic portraiture so let's take that one and generate the image describe is also very easy to use on a mobile phone because you don't need to upload an image to Discord separately you just upload it directly to the command [Music] okay next is slash blend so when I was showing you image pumps you saw that on a mobile phone it's very complicated to upload an image and then copy the URL and then paste it into the text prompt all that becomes very complicated and what if you just want to take two images and combine them and mix them well that's what slash blend comes in with Slash blend you can just give the slash blend command and then let me joinibot will basically ask you to upload up to 5 images and combine them together blend them essentially very easy to use on mobile but a lot less flexible than textbooks this is what we've added image one and image 2. okay look at that crazy stuff that came out of this blend but to be fair we also fed it really crazy images so let's see how it go goes when we upscale it and that's what we get I mean it is a pretty good blend of the two even though it will be the fuel of my nightmares going forward niji is a special version of the mid Johnny engine that allows you to generate more anime style images essentially it was trained on anime images instead of photographs and because of that in general it's much more Japanese cartoon style images me Johnny has ninja version both for version 4 and version 5. it can be fun to generate those images but again it's less flexible than the default engine here is the image with ninji and here's the exact same prompt with V5 right second batch of tips because you love them so much and I love giving you tips so let's say you generate a mid Journey image and the face in the image is completely broken well previously I showed you gfb gun but does a much better algorithm available now it's called Cod former and when you feed it an image it basically upscales it and cleans it up keep in mind that in order to use code former you need to sign in with GitHub it's free but you still need to have an account and then you can just upload the generated image you can upload the general image with a token face and it will fix the face and make it more kind of smoother what if you want to take an image you generated from a journey and you want to print it on a canvas or make it you know an art piece for your living room well the image is generated from I mean Johnny even the journey 5 generates much higher quality images while still too small to print yeah at a high quality and for that we have ESL gun and with ESL gun you can upscale an image and basically four times while I want to say retain some of the image details but reality is that's not retaining that's actually inventing details outside of kind of the generation algorithm then you can use it for free on replicate as well the same way that you can use gfb gun or chord former and in fact he has saw a gun even has Jeff began built in so if it's upscales and the image has faces in it the faces become cleaner and kind of smoother in a way so I'm saving it here and feeling it to real Sargon and submit and within a few seconds I get a very high resolution upscaled version of that image but I can easily print but I hope you enjoyed this video I know it's been a lot and it's been a while since I did a video on generative AIS but I promise I will have more of those coming and there's some very very exciting ones coming I've been traveling and it's been hard to making videos while I'm away from my studio be sure to visit our Discord Community where you can play with the mid join version 5 and now you can play that with blue Velo and other algorithms as well if you haven't seen my video on Instagram pics to picks definitely check it out here as well and show the most amazing results and images you have achieved you know this code and as always see you soon with a new Digital Life Hack and new algorithms including AI algorithms and speaking of algorithms be sure to like And subscribe this video so the algorithms knows to recommend it to more people and until next time bye bye

Channel: Hacking Modern Life

Views: 36,827

Keywords: ai art generator, artificial intelligence, artificial intelligence art, midjourney, ai generated art, ai art generator app, 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, midjourney version 5

Id: DhBuaIW5WsA

Length: 18min 17sec (1097 seconds)

Published: Wed Apr 19 2023