Secret Photorealism Formula in Midjourney | A Detailed Guide

well it's here it hasn't even been a year since mid-journey was released the public and now version 5 is better than anyone's wildest imaginations yes it can do hands yes it can do multiple figures close up mid-ranged and far away yes it can generate a consistent character in different lighting from different angles with different Expressions it can do all of this with beautiful photorealistic imaginative images that hardly take any work at all from a prompting perspective and if that's true what's the point of an advanced mid-earning guide anymore AI has progressed so far that I can teach you photorealism in mid-journey by telling you to type two words before any prompt yes two words you don't believe me well I can prove it the words are photo of you can generate a photo of an ice cream cone melting and dripping through the grates of a sewer drain sorry I need to quit starting with a gross one one of you needs to like go into the comments and just yell at me because otherwise I think this is gonna keep happening we could do a photo of a young man with poofy blonde hair struggling to tie his shoes that image would be a garbled mess in any other model but here it looks great and especially when I only give you a couple seconds to look at it it looks like a photograph all right I'm having fun let's do one more how about we return to the fan favorite my photo of Roger he certainly looks photorealistic and Miles more so than anything I did for him in version 4. look how cute he is while he's asking you to subscribe cool right but what gives you're probably asking yourself why would I make this big long video when the lesson here is that you can take two words and suddenly you're a master of photorealism is this video about to be a waste of time well if you know me you know I'm here to take mid-journey generations of Photography to new depths and unlock skills you hopefully didn't even realize were possible skill in prompt engineering is no longer marked by the ability to generate images which are technically competent and Visually coherent that's because these tools have advanced so far that now anybody can do that much instead prompt engineering is turning into the ability to have fine control over a model's output it's the sense and the patience to gently guide the image AI into creating complex specific and beautiful artworks that otherwise could never be realized in a traditional medium my goal in this video is to give you the tools to do exactly that first things first we need to define a framework so that you can easily keep track of what the words in your prompts are doing to your images I promise you this is cool and not boring and it's also way easier than it used to be in version 4 I needed to break up my prompts with these multi-prompt markers and weights to ensure I was generating exactly what I needed to the reason was that you needed to fight to convince the model to generate something photorealistic but that meant referencing camera gear and settings and if you weren't careful suddenly your images would be full of photography equipment instead of focusing on your subject that meant promscot long complicated and rigid now in version 5 mid-journey has become way smarter and what that means for your explanations of things like lighting and style is well they can be extremely simple consequently so can your prompts when you write a prompt you need a subject you need a style if you care what the background is you need a background if you care how the lighting look you can explain the lighting same thing for the colors in your image same thing for the mood and if you have done that in the eyes of mid-journey you have fully defined your artwork and as we know a fully defined artwork means mid-journey's model no longer needs to inject prompt engineering and change the image for you make sense let's do it pick a subject we'll call her Becca she is a young lady with long bleach blonde hair and with a joyful round face sorry my bad the problem should probably be a photo of a young lady with long bleach blonde hair and with a joyful round face now a background she can be standing in front of a crumbling wall and a style I'll call this Urban Chic that worked it's good but I don't think it changes the image quite enough so I'll add a few details to this prompt and solidify the Style again you don't need nearly as many details as you used to version 5 is very smart and it understands the style you're talking about just be specific and ta-da it looks great here I've also written another styling prompt to define the lighting and mood so here we can add my prompt for Twilight and wow suddenly the image looks gorgeous what's cool is that in this format we can break each of these down into slash prefer options it's for subject background style and atmosphere this way we get to Define our subject once and then try generating them with a ton of different and compelling combinations we can take Becca and then create the same portrait at Sunset instead of Twilight or replace Urban Chic with high fashion we can also do a fantasy portrait of her under what I call God Rays the point is you get to have tons of variations to mix and match all of these attributes until you land on something you love you might have noticed that each of these dash dash some things are slash prefer option set commands each with a bunch of text associated with them it would take you a monstrous amount of time to copy all these down for my video luckily I'm your favorite YouTuber and I've done something amazing just as before in the description of this video I am posting a paste bin with all of the slash prefer options I use in this video plus more than two dozen extra so if you add them to your mid-journey account by pasting the commands in you can use them as well they aren't complicated but they do give you an excellent starting point to creating refined images and then you actually get another superpower Mad Lib style you can actually write a bunch of prompts that look like this slash prefer options subject background High fashion Sunset bokeh and then very simply all you have to do is type out your prefer option command for your subject and your background and then you can try them in a bunch of different and refined Styles simply by re-pasting a bunch of commands which I also put in the paste bin so let's get to try to let's set our subject slash prefer option subject then click tab into the value a photo of a handsome owl person wearing glasses and a trench coat then we need to add a reasonable background how about a slash prefer option set option background value in a dark and spooky forest with shadows peering in from behind the trees now I can grab these handy prompts and very simply run them all suddenly you have a style sheet full of refined images in a bunch of different styles colors lightings moods and perspectives you can change the option for stash subject and Dash Dash background then do it all over again with a completely different character straight out of your imagination but we're not done my paste spin is just a start hold on to something and or sit down because I need to apologize I haven't been completely honest with you you know the 40 odd style hints I put in this astounding paste bin for you the one you were so grateful for that you already click the like button without being asked thank you by the way well I didn't write any of them they came to directly out of chat GPT and as a special treat for watching this far I'll tell you that within the first Pace bin is a link to a second pay spin where I put the prompts I use to generate all of them you're welcome with these tools available to you and a bit of practice you will become a master at augmenting your prompts so that you can turn anything you imagine into something mind-blowingly awesome and to finally turn you into a master of mid-journey I just need to leave you with three final tips number one if you want to create a consistent character you need to be specific when writing a prompt for your subject include all the details mid-journey will need to understand how that character looks if you make your subject a cute guy he will be different in every single scene you put him in but if you add a ton of details and instead write Alec a young Emo looking guy with flowing long blonde hair a nose ring chiseled Jaw and a wistful gaze well that character is going to look much more similar when you generate him again and again consistency is complicated and I could do a whole video about generating consistent characters but prompts like this will give you a much better shot at it than anything else ready for the next one tip number two never trust the order mid-journey is smarter than it used to be but it doesn't know everything mid-journey still has problems with prompts that involve complex relationships take this a white ball wearing a green Top Hat rolling on a black ramp being chased by a purple pyramid the colors the objects and their relationships are essentially randomized mid-journey doesn't know what to do with it and that's how it's designed in some cases it's a good thing it means you can switch the order of these tags I've given you and not actually affect the outcome of the prompt but it does cause cost you the ability to specifically Define what part of the clown should be what color number three this landscape is constantly evolving in my opinion mid-journey has been the best source of image generation basically since it came out and since I made a video on it but if you want to be using these tools to create the best possible images you need to stay aware of how AI is changing stable diffusion has really cool developments all the time there are free models popping up seemingly every week Microsoft designer Adobe Firefly there is a lot of news to cover and it's a lot of work to stay up to date now you might think that this final tip is just a cleverly disguised plug to subscribe to me but I have to break it to you all that news I don't cover it there are some great channels that do Matt wolf all about AI Matt vid Pro are all excellent at sharing timely updates about this amazing Tech if you want to stay up to date following them will be a great way to turn yourself from an mature dabbling with me journey into an informed professional who is aware of all the goings-on in this space my goal with this channel is much more to help you understand and think about these Concepts more deeply hopefully I've been able to demystify some of the things that help you approach AI art from a position of agency and freedom I hope you agree that mid-journey version 5 is a fantastic tool that is going to reshape The Conversation Over what's even possible from AI but I have to add that this revolutionary Improvement is nothing compared to what happened the day GPT 4 was released if you don't believe me you'll just have to make sure you watch my next video see you there

Channel: Glibatree

Views: 95,692

Keywords: GPT-3, AI Demo, AI Story Telling, AI Art, Midjourney, ChatGPT

Id: IGHpWh09JXQ

Length: 12min 30sec (750 seconds)

Published: Mon Apr 10 2023