Transcript for:
Mid Journey Image Generator Guide

[Music] there's a lot of great image generators  now but mid Journey still wins for almost every   use case it has changed a lot over the past  year the largest change being the launch of   the website it's a gigantic improvement over  Discord for usability Aesthetics inspiration   just all around better everything so I've  broken everything down with visualizations   to help understand every parameter systems for  prompting resources style reference explorers   tons of examples some more unique workflows and  lesser used techniques everything you need to   master mid Journey as a complete beginner or a  longtime user and once you get the hang of mid   Journey it feels like magic [Music] let's Jump  Right In here and get oriented with the site   I'm not going to talk about anything in Discord  this will be solely focused on the site because   it's just so much better although everything will  also apply to Discord if you want to use it there   as of the filming of this video you need to have  generated at least 100 images in Discord before   you have access to the site we start out on the  explore tab where you'll see images other people   have generated with different ways to sort them  it is the best spot for Endless inspiration using   the search box you can search through by keyword  to find things for inspiration or to use as style   and image references when you click on an image  it will show the full prompt and parameters they   used click here to use the prompt or here to  use it as a style or image reference you can   also do that by dragging it up to the prompt bar  and you can do that with multiple images to blend them this mushroom has these little lights that  made me think of a carousel so I ran with that as   a style and image reference and came out with  this we'll Deep dive into all of that later a   really cool feature is when you hover over an  image if you click the little magnifying glass   it will pull up other similar images I make a  habit of liking any images that I really like   either the concept or style then I can sort  by liked images and see them all you can find   amazing inspiration this way to never run out of  ideas as you can see I like a lot of weird stuff   sometimes it's fun to just find the weirdest  picture click the magnifying glass then find   the weirdest picture out of those and repeat  until you have nightmares or maybe that's just   me in the Middle where it says imagine we  have the prompt bar to generate just type   in here and hit enter I'll start with a single  word imagination send it a quick tip is you can   hit the up Arrow to get your prompt back into the  box this drop down on the right will open up the   parameters I'll cover all of those in a second  but they're easy to use just slide them around   and then generate again this number popped up here  that shows how many you currently have generated   that's on the create tab which I can open  and watch them generate and see the prompts   and parameters on the side then this archive tab  shows all your previous generations as thumbnails   so it's easier to look through your history and  to organize there's different ways to sort and   view these to just images you've upscaled only  landscape or portrait or older versions of mid   Journey things like that you can change the  layout to full so you see the entire images   instead of just the square thumbnails or you can  change the size and you can search through by date   if you hover on the bar on the right so you don't  have to scroll through thousands of images to go   back I highly recommend getting into the habit  of liking good Generations you make so you can   find them easier later you can create folders by  either manually selecting images hold down shift   then click and drag to select or you can also  create smart folders so say I want every logo   I've made I just type the word logo then mid  Journey instantly creates a folder with every   time I've used the word logo in a prompt I'll also  mention that down near the bottom you can switch   between light and dark mode or just leave it on  dark mode all the time like a normal person now   next to create I can see it's finished averages  around 30 seconds to complete so I'll click that   and these images look amazing and now I can go  open up any of these generations and there's a   whole list of options I can do with these from  here vary upscale reframe repaint a lot of that   is pretty intuitive how it works but we'll dive  deeper into that as we go through these images are   awesome these one-word prompts are really fun to  see how mid Journey visualizes a concept you have   no idea what's going to come out on the other side  but it's fun I've done a bunch of these [Music] [Music] you can prompt with Emojis [Music] too so yeah  type a Word get a cool image that's great but   we want to have more control the main ways  to control the output are through parameters   prompting and references you can do all of that  right up here it's incredible just how much as   possible all within this little area clicking  here will let you upload images for image style   and character references or drag them here like I  showed before the icon on the right gets me back   to the parameters so in short parameters change  the way mid Journey interprets your prompt that   lets you explore and fine-tune an infinite amount  of aesthetic directions I'll start with just the   parameters that are here when you click this  dropdown these are your main settings when you   change one it will apply to every prompt going  forward until you change it again under image   size we've got portrait square and Landscape  or you can drag the slider to adjust the aspect   ratio they make it really easy there's tons of  reasons for using any of these not only based   on where you may be posting the images but it also  influences the composition and feel of them if you   want an aspect ratio that isn't here it doesn't  look like it but you can actually type right in   here to replace the numbers with what you want now  I've got one four in there maybe I need a bookmark   giraffe on roller skates or it could flip that for  a panorama of a concert hall or you can take that   even more extreme if you want like this layout for  an 8bit sidescroller adventure game you'll see it   when I hit up to get the prompt back it shows the  aspect ratio parameter at the end every parameter   can be typed manually they will all start with  a dash dash and they always go at the end of   your prompt also anytime you do type a parameter  manually it will override what you have in this   drop down I have all of these parameters as well  as all the others will cover in one cheat sheet   that you can use as a reference that's down  in the description along with a ton of other   mid-journey resources there's also a free chat  GPT resource bundle provided by HubSpot there's   a total of five PDFs that go in-depth on how you  can utilize chat GPT in your career to get ahead   solve problems or save time my favorite is called  how to supercharge your workday with chat GPT it   covers specific examples of how chat GPT can be  used in various Industries sales and marketing   project management enhanced decision-making and  problem solving time management and organization   it walks through step by step with different tips  and even has a section titled 100 ways to try chat   GPT today with 100 sample prompts you can use and  modify no matter what career you have there's sure   to be a bunch in there that apply and that's  just one of the resources in the bundle which   again is completely free just use the link in the  description to go download that I have been happy   to partner with HubSpot to provide free resources  to the people that watch this channel moving on   to Aesthetics we've got stylization weirdness and  variety stylization adjusts how artistic the image   is and by artistic that means favoring aesthetic  colors composition and forms so here's a good one   to demonstrate that a child's drawing of a cat the  low stylization definitely looks more like a child   drew it on the high end it favors an artistic  output which looks great but you would have a   very talented child if they're drawing that here's  examples on some more types of prompts so a coffee   shop logo low stylization is more minimalist the  higher end gets more complex with this holographic   car these first shots look pretty realistic as it  goes up it gets more saturated and contrasty just   looks more staged and edited in the selfies with  Bigfoot the low stylize has poorly framed photos   that feel more authentic as we go up it's more  like they're taking perfect photos for Instagram   and that totally gives away that it's AI the  real big foot would never do that in a more   artistic style this abstract Tree of Life As  you move up it adds more contrast and depth or   with a portrait shot like this one it starts out  showing a lot of details in the face then starts   to move younger and more of that filtered look  so the low end follows the prompt more closely   the higher end puts Aesthetics first the default  of 100 is a pretty good balance between those I   rarely take this to the high end but I do actually  like to go between these values particularly on   the low end like 50 or 75 to do that type Dash  D- stylize space 50 you can also just type D-   s either works zero usually doesn't look very  good unless you have a really detailed prompt   a lot of the time I stay in the range of probably  40 to 200 weirdness is to explore unconventional   Aesthetics it's really fun it can lead to some  very unique outcomes I usually keep this on the   lower end but there is definitely times when  you want to explore and get more unconventional   here's a simple prompt and apple on a table so  starts with a very literal and simple image then   as I increase the weirdness I start to get more  unique compositions angles and perspectives you   would never expect how about clown sitting at  a bus stop that might get a little weirder all   right these first ones are a traditional looking  clown two of them are just Pennywise and they're   at normal looking bus stops at 600 we start  getting one where the clown is off to the side   on the ground with his foot sticking out or this  one on the bottom right is not like any bus stop   I've ever seen a lot more unusual at 12 00 I don't  even want to open this one up to figure out what's   going on here we' got more different poses and  angles then it tests out some black and white and   long exposures I ran a few at 3,000 we've got some  selective coloring and just all sorts of weird bus   stops and interesting poses and compositions this  is a pretty mild prompt if you go with something   more out there the weird parameter can definitely  take you to some strange places I want to show one   with a more specific prompt like this one which  is a good example of how much detail you can add   into a prompt in mid Journey it nailed every bit  of this but when I increase the weirdness a little   bit it quickly goes from beautiful Framing and  people and poses to more unusual but it already   has lost the setting of a fine dining restaurant  and the lighting and color grading it retains most   of the rest of the details through all of these  but it defaults to less attractive people and   poses like while they're in the middle of eating  or looking bored or annoyed with each other if you   want a weird image that is still conventionally  attractive you can try increasing the styliz   parameter variety is how varied the outputs  are the lower values will have more consistent   results and then higher values will lead to a  wider range of compositions now I might bump   this up when I'm exploring then when I start  honing in on something I want and I'm getting   specific with my prompt I'll start lowering that  to get consistency and when you send this it's   actually called the chaos parameter this one's  pretty straightforward so here's just a couple   examples this is all fairly minor variations then  we start to get a wider range of interpretations   so to summarize what these are and how they're  different from each other stylize controls how   strongly mid Journey's default aesthetic is  applied weird controls how unusual an image   is compared to previous mid Journey images chaos  controls how diverse the initial grid images are   from each other they're all different but all  work together and influence each other as well   they make it just really easy to use with these  sliders or you can type them manually to get   numbers in between for mode we have standard  or raw raw gives you more control over your   image and has less automatic beautification  applied basically it listens more closely to   your words and interprets them more literally is  especially useful for photographic Styles if you   want images that are as realistic as possible  that will be style raw and low stylized value   side by side comparisons are the best way  to show this one I very often will just try   both so here's against different photographic  prompts now here with some more Artistic Styles for versions version six is the default right now  which is currently the latest version but you can   select previous versions of mid Journey if you'd  like to all the way back to version one six has   the best prompt coherence and knowledge and is  generally the most advanced and photo realistic   it does tend to add a lot more details too which  you may not want 5.2 was also great and can be   a little easier to get results by just typing  in keywords there's also nii which was trained   to have a much wider knowledge of anime and  illustration Style but it can give interesting   results for other Styles as well so those are  the main ones to use that generally speaking   will give the highest quality you can go back  further those images won't be as high quality   but can be fun for an experiment personalize will  use your own personal aesthetic preferences to   influence the generation you have to rank  image pairs to train the algorithm on your preferences I love what my personalization does to  the images but this is a bigger topic that needed   a dedicated SE ction I'll go over this closer to  the end it will just be turned off until then for   Speed I'll just run a prompt at each of these  speeds and let it run in real time as I explain   it the default for Speed is fast that's what  I use almost all of the time there is a cap on   GPU time for every month depending on which plan  you have it's roughly 1 minute of GPU time per   generation that means the basic plan is around  200 Generations per month on fast mode standard   is around 900 Pro 1,800 and mega 3600 you also  have the option to buy more GPU time if you're   run out also if you're on the standard plan and  up you get unlimited Generations in relax mode   they do take longer to generate roughly 1 to 10  minutes depending on the time of day so that can   be a good option especially if you run out of  minutes and don't want to buy more then there's   also turbo mode which generates four times faster  but uses twice as many subscription GPU minutes   as a fast mode job you can earn free GPU hours by  rating images under the tasks tab this is also how   you train that personalization into mid Journey  the top 2,000 rankers per day will get a free f   fast hour this used to take only around like 5  minutes of ranking to earn a fast hour but now   that they've added personalization people are  using it a lot more right now so it will most   likely take longer fast and turbo We Done Quick  but this relax mode is still generating there's   two manual parameters I want to mention here  before moving on to prompting the first is the   no parameter for if there's something specific you  don't want in your image so I want a bowl of fruit   with no bananas dash dash no space bananas send  that and it will come back with no ban bananas   so that is a really useful one and the other I'll  mention is seed so if you don't know mid journey   and other diffusion models they start by creating  an initial field of visual noise then they refine   that noise then gradually transform that into  the final image so each time you generate it   uses a random seed number to generate the initial  field of noise to get the seed number go down to   more over here then click copy then seed now I'll  run the same prompt with-- seed space then paste   that number now here's what that looked like  with and without the seed now here's that same   seed with completely different prompts it does  influence the scene but not in a way that's very   predictable so I used the same seed for a lot of  those comparisons like the child's drawing of a   cat that's why the composition was so similar  across those if I didn't specify it would have   used a random seed each time and they would have  looked a lot different like this you know still   makes sense what the styliz parameter was doing  but it's not as directec of a comparison all right   one other thing I'll mention here is permutations  it's not a parameter but you can use permutations   to run different variations of your parameters  or prompt without having to type each of them   out I did that for most of these examples to run  at each of the different stylized values use a   curly bracket then each value separated by commas  then close the curly bracket and when I send that   you can see it's running the same prompt with  each of those styliz values you can do the same   thing with words in your prompt like if I want  to test out different colors or camera angles   just anything I'll demo more examples using that  as we go through I know that all seems like a a   lot if it's the first time you're hearing it but  it all gets pretty intuitive once you understand   them and have used them just a few times there are  more parameters not in this box some based around   references which are really powerful for guiding  the scene but I want to switch it up for a bit and   cover prompting before we get into that prompting  is a big topic so it's a long section we've got   all those parameters but they don't do anything  unless they have a prompt to interpret the prompt   is still the primary driver although that will  change some when we get into reference images and   it's important to not that there's no right way to  prompt you will never find the perfect prompt it   always takes some experimentation and guiding and  bouts of serendipity to get something you like but   having a structure to prompt with can help a lot  the simple approach is this scene subject details   setting details style details I guess you could  call it the four s's scene subject setting style   and by scene I mean an overview of the scene using  just generic archetypes and keywords all right so   I'll start with a scene and then just build it out  to show how it changes and anthropomorphic bar and   owl exploring a dimly lit abandoned Temple I'll  do 169 and style raw all right that looks good   but not at all what I'm going for so I'll add some  details about the owl the owl is wearing CT robes   and a hood and holding a burning torch and calling  back to the term you used in the first sentence   in this case the owl will help mid Journey know  what you're describing instead of thinking it's   a new character same thing when describing the  setting so I will call back with the temple has   Dusty relics and Faded murals on the walls dark  and mystical atmosphere with soft Eerie lighting   this already is pretty much exactly what I was  looking for you I had this kind of cinematic   dark fantasy vibe like pan Labyrinth or something  along those lines if I wanted to describe that I   could use the director's name in the style of gyo  Del Toro or find the type of camera or film grain   it was shot on I usually have better results from  using film grains rather than camera names I mean   which makes sense it looks like there were three  film stocks used I'll go with 35 mm Kodak Vision   2 500t 5218 and that looks a lot more consistent  I could refine that a little more now that I have   pan labyrinth in my head instead of soft Eerie  lighting I'll switch to Loki lighting and Deep   Shadows and to refine that I'd start experimenting  with parameters from here like up and down in   stylize you know 50 75 200 and 300 then try  standard instead of raw then I'd just iterate   until it's perfect but this looks great here I  think you get the idea I'll do another one in a   photo realistic style this time with multiple  people all human candid shot of three friends   at a cafe I'll send that then hit up to get the  prompt pack and add some details I'll call back   with the friend on the left is an African-American  man wearing a blue button-up shirt then again I'll   say the friend in the middle is a blonde Caucasian  woman wearing a green blouse the friend on the   right is an Asian woman wearing a sundress with a  floral pattern iPhone photo I used the friend for   each of those to make sure it knew I was referring  to the original friends this is more important   when you have more different archetypes like if  I used a man a woman and a dog in the scene you   can see what the defaults for Mid Journey tend to  look like if you don't describe the subject this   one looks great it nailed it in almost every one  of them and like always I can test out different   styliz values zero usually looks pretty rough but  these others below have them looking at each other   more like they're having a conversation which  is what I was intending by saying candid shot   although it would have helped to write that out  in the prompt and the higher up you go the more   posed and filtered they look and lower cut  the shirts get apparently when you use style   raw and words like candid and iPhone photo it's  actually pretty insane how realistic they can look [Music] when I say details that's a really broad  word there's a huge range of things that   can include depending on what you're trying  to create mid Journey breaks down words and   phrases into smaller pieces called tokens so it  takes those tokens Compares them to its training   data to inform it how to generate the image some  words will have more impact than others the ones   with the biggest and most interesting impact  are often called Power tokens I made a list   of some of of my favorite Power tokens I've  come across they can each have a big impact   on their own and of course you can use any of  them in combination and see where it takes you [Music] [Music] now a lot of this so far has been based  off of Photography terms but these can inform   the scene across mediums you I could have done  this in a different style pretty easily remove   the previous style and just adding a couple  words into the summary part of this prompt   is enough then I could guide it more in  the style details and depending on what   style you're aiming for a good practice is to  learn about that style having a vocabulary in   a given medium is really helpful and overall if  you don't already have that knowledge it's just   a good excuse to learn about various art  forms and artists and techniques there's   a lot of places you can go to help learn about  this stuff by far the best mid Journey focused   resource I have ever come across is mid library.  it's an amazing way to explore Styles and artists   and artistic techniques and how mid Journey  interprets them it's an invaluable resource   and an incredible way to get inspiration you can  go to all styles and click Advanced filters for   a bunch of ways to filter or the classifier tab  has them categorized in various ways techniques   fashion design ERS Architects filmmakers  illustrators also has this area for artistic [Music] techniques or this one for  General modifiers which is more   like [Music] keywords there's a bunch of  really insightful guides here as well like   this one about mimetic style emulation  and the process to learn about a style   in order to recreate it rather than typing an  artist's name I'm not affiliated with site or   anything by the way I've just found it extremely  helpful and something that you can do with mid   library is instead of using the words you  can pull these into mid journey to use as   style references I'll be getting to those  in a little bit it's a whole different way   of prompting that's done a lot more visually  mid library is a great spot for that [Music] too I ran just the prompt Dragon then added some  of the coolest style words I found on Mid Library [Music] another way to get help with styles to use  and words for your prompt is to upload an   image then click the little I icon and  it will describe the image for you so   it will give the subject known artists  with a similar style and descriptors   you can go through and find interesting  words to try out and see how they affect   the image that can be a good way to get help  describing a style if you're struggling with that that basic prompting structure applies across  most categories but there will be more specific   terms you may need for things like architecture  fashion or product photography those are a little   more Niche but easy to adapt to one I want to  dive deeper on since it's pretty common but   very different is Vector art you know for Logos  branding or print products I'll start with a logo   generally I keep the aspect ratio as a square for  Logos I'm just saying coffee shop logo would work   there's a lot of specific words that can help  some of my favorites are minimalist abstract   brandmark and geometric or a combination of those  you could do something more specific to your brand   like Boutique psychedelic retro or maybe they're  also a bookstore then of course experiment with   the styliz parameter going up will affect things  like minimalist or brandmark a little bit they'll   still get good results at high styliz values but  something like psychedelic will have a big effect   and we'll usually get too complex for a logo  past 200 and logos are one that can be pretty   interesting to try out famous logo designers  names with like Milton Glazer Paul Rand Carolyn   Davidson and Ivan CHF you can upload your Brand's  color palette as a style reference if you have   them as well I can add text to any of these two  by adding quotation marks around the company name   Kevin's coffee the text is perfect on this one  I'll get into more specifics around generating   text and how to fix these in just a minute but how  did it know to put a guy with a beard on this one   a lot of that will also apply to Vector art if  you're making things for printing on t-shirts   mugs or whatever it might be some additional words  to use would be Vector art Vector graphic graphic   design then things like icon emblem shirt design  vyl sticker or clip art can be useful even if   you're not specifically making it for one of those  silhouette or outline can be cool or some from the   logos like geometric shapes abstract minimalist  then for something on the complex side you could   add complex or a wordlike illustration try out  color palettes any combination of those things   there's a couple ways to convert these into vector  graphics Adobe Express has a free converter just   use the PNG to SVG it works pretty well I used to  use vectorizer doai it is better than the Adobe   converter but it's not free anymore if you're  doing this a lot it could be worth it to pay   for that then bring it into Adobe Illustrator or  whatever program to edit it further sometimes the   results from these converters are weird to work  with especially if you use a complex image back   to the topic of text you can generate text in  mid Journey but it is one of the more difficult   things to get right consistently to get text just  add quotation marks around the text text you want   it is case sensitive and can also do special  characters things that normally have text on   them are a lot easier to get results with like  paper license plates theater signs Billboards   a lot of the time you'll get it first try on the  easier things but for even slightly complex ideas   it is not consistent so you'll usually need  to roll a lot this one wasn't bad two of these   are good if something is close like this one try  running variations and just re-rolling the closest   until you get it if there's minor issues you can  try inpainting to add or remove letters or just   imp paint over the whole text I got at least one  usable one with each option the subtle variation   strong variations and in painting it's actually  very good at matching text to a style after the   fact so I can grab one of these logos and select  the area where I want my brand name to be I'll   just use the word coffee and it matched this  style perfectly with the colors and gradient   and a pretty decent font so now I don't have  to bring that into Photoshop or illustrator and   the text it's ready to go before we move on  from prompting here's a few more just rapid   fire top prompting tips I'll actually start with  a few from the mid Journey documentation then do   a bunch of my own the simpler you can make your  phrasing the better instead of show me a picture   of lots of blooming California poppies make  them bright vibrant orange and draw them in   an illustrated Style with colored pencils instead  of that try bright orange California poppies drawn   with colored pencils mid Journey breaks a prompt  into smaller pieces called tokens and this first   option gives a lot of unnecessary or imprecise  tokens like show picture Lots the second option   is only the words that are intended to direct  The Prompt it's much more precise word choice   is important the more specific you can be the  better instead of big try High gigantic enormous   plural words leave a lot to chance if you say cats  that could end up in 3 10 or 100 specifying three   cats would be much more helpful focus on what you  want instead of what you don't want prompting a   birthday party with no cake will usually end up  with a cake if you want to make sure something   isn't there use the no parameter specify what  you want anything you don't specify will let   mid Journey fill in the details so include every  element that is important to you whether that's   the subject medium environment lighting color  mood composition it could be a lot more or a lot   less if you're trying for a full body shot specify  the shoes M Journey cuts off the bottom a lot of   the time if you specify the shoes it will almost  always generate the full body the easiest way to   get something cinematic is to add either cinematic  still or thir mm cinematic if you're struggling   to figure out how to prompt for something try  searching on the explore tab often you'll find   inspiration or a prompt to build off try switching  between style raw and standard that little change   can make a big difference if you want some random  images forget everything I just went over and add   emojis into the prompt box and see what happens on  that note just generally experiment again there is   no perfect way to prompt utilize these tips and  also try the opposite just experiment you never   know where it might take you now that we have  some cool images there's some next steps you   can take one when you click on one I've shown  a couple of these already but I'll dive into   them further we have very subtle and strong the  very subtle will give four new images that have   similar composition and colors to the original but  with some subtle changes then the very strong will   have more differences in the composition colors  and details almost every time I have an image   I like or once I'm getting close I'll run these  to get some more options you can upscale subtle   and creative both of these will double the size  of your image subtle will keep the details very   similar to the original then creative will add  new details to the image you can experiment with   each to see which works best for any given image  both the upscaler do an amazing job if you have an   image where the face is a little off or distorted  Sometimes using a creative upscale will fix it rerun will rerun the exact same prompt as is  with all the parameters included reframe has a   few functions within it you can quickly change the  aspect ratio under reframe then I can click Start   or end to move the image into a different spot and  it will generate the missing areas under Zoom you   can zoom out all the way up to 2x while you're  on reframe or Zoom you can click edit prompt and   change anything about the prompt to influence  what it generates in those areas it's expanding   into I could pan and add new people or objects  into a shot or as you zoom out you can zoom out   into entirely different scenes there's some really  interesting ways that can be used there's an old   video I made where I uploaded an image of my desk  setup and zoomed out and panned into all sorts of   different scenes then edit it all together I  can change up the scene with every Zoom or I   could pan to different locations I can do that  in any direction I'd like and there's even more   possibilities to throw some other AI tools on  top maybe even add a character this has been a   lot of fun I've also seen the door Brothers  make some really creative music videos using that also you can change the prompt without using  the reframe or Zoom how about a lobster body what   that does is actually called remix strong if I  click this more options button then select remix   it shows that I've already ran a remix strong  because that's what I just did but there is also   this subtle option here I'll use that as well for  a comparison when I click it the prompt pops up at   the top and I can change it I do actually like to  use remix subtle a lot it's the best way to make   changes while retaining a very similar composition  to the original super good for just quickly   iterating on things that got way creepier than I  was expecting with the lobster so I'll switch to   something else for the other tools sometimes I'll  really like the image but there's one part that's   a little off about it that definitely happens  repaint is inpainting or selecting a specific   part of the image to change instead of the entire  thing let's say I want to have open eyes here   select with either the rectangle or the lasso tool  you can leave the prompt or change it I'll WR what   I want for this one and there it is this comes in  handy a lot like if you've got some hands that are   off use repaint and it should fix it eventually  sometimes it can take a few tries or replace a   character you can even replace a face with a face  of your choice using the character reference which   we'll get to soon references help in many ways to  get results you want faster and more easily the   overview of these and how they differ is a good  place to start image prompt pulls in more of the   structure of the image the subjects forms and  motifs style reference focuses on the aesthetic   like the lighting colors more like the vibe of  the image character reference looks specifically   at a character and will'll try to match the face  most importantly then depending on how you adjust   the weight it will also pull in clothing colors  and style even more simplified would be structure   aesthetic character a lot is now done through  references rather than prompting overall for   me it's a lot more fun to blend and direct things  with images rather than language just feels more   intuitive and also faster but words are still the  best for positioning and describing subjects in a   scene so I'll cover how each of these work first  before how to use them all together for each of   these all you have to do is drag an image in  I'll use this little mushroom again then select   which type of reference you want it to be if  you want the same image to be multiple types   of references hold shift and click the others  for image prompt it's the the little picture   that's what it defaults to I'll try to change this  style to a pencil sketch of a mushroom that looks   great it's a very similar composition and type of  mushroom but in a completely different style this   looks good but we can also adjust how strongly the  image influences the output with the image weight   parameter D- IW then a number between 0 and 3 four  version 6 between 0 and two four version 5 so I'll   run a permutation with the weights. 25.5 75 1 2  and 3 went down at 25 is a basic pencil sketch   0.5 has that smaller mushroom more consistently as  we go up more we start getting more color and more   of the forest but straying a bit from the pencil  sketch and you don't just have to try switching   style I'll use this same image reference and just  type wizard at the default of one it gets me a   similar composition but with a wizard and the  hat is slightly mushroom shaped and has these   dots on it there's also more mushrooms within the  image and for reference here's just running the   prompt wizard without an image reference it's  a big difference here's that different weight   below one looks great still when you go up to  two it doesn't look like a wizard with mushroom   characteristics more like a mushroom with a few  wizard characteristics like the more flowy beard   likee stem and the cap a little more shaped like  a wizard hat the cap with the wizard hat shape is   actually pretty awesome you can also use multiple  image references it can be fun to just browse the   explore Tab and find random cool images to use  here's a cool double exposure I'll use that then   this looks like a version of the girl with pearl  earrings I'll use that and send it let's see what   we got that actually looks pretty cool I'll jump  to my likes Tab and just mix up a bunch of random   ones together and see what happens doing this with  a specific goal in mind is great but sometimes   just throwing random images into the slot machine  can be cool too not necessarily to end up with a   finished image but something cool to work with  further or get inspiration from here's what I   got some of these are cool nothing spectacular you  could add a prompt into this process as well to   guide it more now I'll go back to that mushroom  image but switch this to a style reference   instead of an image prompt I'll prompt it with  wizard again and submit this time it retains   essentially none of the original composition in  structure but uses the same colors and Vibe as a   comparison here's the default style reference and  default image prompt next to each other then just   to reinforce what the style reference is doing  I will switch up the prompt and we can see the   scene changes but the style Remains the Same and  just like with image prompts we can control how   strongly the style is applied by using-- S SW  for style weight then a number between zero and   a th000 here's that same prompt at different style  weight in this case I prefer it in the lower range   but it will vary from image to [Music] image again  you can bring in multiple images to use as style   references to blend those Styles just using the  prompt wizard I'll throw a bunch of random Styles   together do single references some with two some  with four we'll see where it goes it's really fun   and starts to feel more like mixing paints on  a palette and some of these look just amazing [Music]   this is so powerful and you can combine  style references with image prompts as   well and again to use a single image as  both hold down shift and click the other   that's what I did to get this mushroom carousel  there was some remixing and in painting as well in addition to these image-based style references  there's also unique codes assigned to styles that   mid Journey can create you can use that to  have consistent style applied across images   that also gets extremely powerful I have links  to some databases where people have been posting   their curated Styles you can use that's coming  up soon but first let's move on to this last   one here character reference so this lets you  create consistent characters and this was the   most requested feature in mid Journey ever until  they finally implemented it it works best with   characters you generated within mid Journey but  you can use non- mid Journey images too I'll demo   both I'll just pick a random character from one of  the prompts earlier this one has some uniqueness   drag it up to the prompt bar then switch it to  a character reference now I'll run a permutation   woman sitting in a cafe walking through a forest  playing chess underwater hiking cyber Punk Warrior   exploring a cave and these look great we can  see her across all these different scenes it   made an effort to match her clothing as well  there is a character weight parameter D- CW   that adjusts how much it tries to match  the default of 100 tries to match the Face   clothing and accessories lowering it all the way  to zero will only match the face and the clothes   are rarely perfect so a lot of the time I have  this at zero as you can tell from these it will   work across different scenes Styles and camera  angles it also works with different emotions if   you prompt for them all right now we've got that  with a mid Journey generated character but how   about an image of me this works sort of just not  as well so I'll try to convert this into a couple   Styles it does an okay job definitely not perfect  with enough rerolls you can get pretty close   you'll never get it exact but you can get a lot  closer there is also the ability to inpaint with   character references if you get a generation where  the face is off or if you want multiple character   references in one image you can add them after the  fact so this is a little weird to do right now I'm   sure they'll make this easier at some point but to  use character reference within repaint you need a   URL for the image I can go to the picture of this  woman I've been using rightclick then copy image   URL now I'll go to repaint on one of these images  from before with the group of friends then select   the woman in the middle I'll change the prompt and  just write the woman in the middle is a cockasian   with red hair then-- CF space paste that image  address and it's ready to send on the generation   we can see it has replace the character  and matches the lighting and style really well if I wanted I could replace these others as  well you can also experiment with using things   other than a character like uploading a texture  gradient or even just a color this is super hit   and miss I'll show some Explorations I've  done getting a guitar in different textures   works pretty well it can get the texture almost  perfectly which is amazing and really difficult   to do in any other way so if there's any reason  you want a very specific texture on something and   depending on the character weight you use it may  bleed into other parts of the scene using just a   color can work to get a very particular Hue onto  something but definitely can run into issues with   bleeding into other parts of the scene depending  on how you use it this worked by in painting as   well this one I thought was pretty cool I wanted  a man with a cube head instead of describing the   cube I could use a cube as a character reference  and it worked really well I also tried seeing if I   could replace a guitar so if I had someone playing  a guitar across multiple scenes the guitar would   remain the same it didn't really work I could get  a new guitar with a similar style which is helpful   but not as perfect of a match as I'd hoped better  than not doing it at all though that's been a lot   of fun experiment see what happens you can  combine all of these reference features I'll   use this character again but I want it in this  style with a similar composition to this image   I'll change the weights of some of those and send  that you can see how much control that gives me   I've been primarily moving towards prompting with  images as much as possible now on to one of my new   favorite parts of mid Journey instead of using  images to give context to the style another way   to do this is with style reference seeds which  is a code for a style mid Journey knows how to   generate and will will give a consistent style  across Generations this is where things get the   most fun there's some databases all link to where  people categorize some of the best srfs they've   found I'll go through those in a second you do D-  srf space then type a number between zero and 4   billion you can also type-- srf space random and  mid Journey will pick a random number for you in   the output you can see it replaced that srf random  with this long number that's the style reference   seed these are two random Styles they look pretty  cool to get more I'll add that same thing srf   random then I'll add a parameter that I haven't  mentioned yet D- repeat 10 repeat runs your prompt   however many times you specify with everything  the same prompt parameters and references but   in this case since I used srf random it will  choose a different random seed each time so I   can explor a wide range of Aesthetics there's some  really cool ones here some weird ones too [Music]   so to drive home what this is doing I will  copy that srf and use it with a completely   different prompt woman walking through  a forest copy the srf now cityscape how   about a comic book superhero and howow  flying I'll do that with a few of these   srfs this is so powerful for getting  consistent Styles across Generations [Music] and just like with the visual style references you  can use the style weight parameter for these with   a number from 0 to 1,000 100 being the default  you'll have to find the balance of how strong the   style is and how much you're trying to influence  it with your prompt and then what you want to show   through each srf will act a little differently  generally you can use them across different   mediums by prompting but some of them will be very  strong Styles and harder to guide in different   directions so that's how they work they're super  fun but it's a complete slot machine to do it this   way and takes time to find really good ones people  have been calling that mining for srfs the process   is fun but can burn a lot of fast hours there's  a few people that I've found mining for srfs   and indexing them I have three pretty solid ones  I'll link to all of them below the first is Ali   Jewels she posts some really cool ones on X and  just a lot of great mid Journey related content   she's a great follow she's also been building  out this visual index of style references she   did recently switch it to paid to view the full  thing but there's still hundreds to explore for free this one looks cool I will copy  that and run it with wizard I'll just   keep using wizard as my Benchmark  I'll just run a bunch of these [Music] now this one is made by Wade McMaster with  Creator impact I don't know how many are here   but there's a lot so I'll grab a few of those  and then here's another one from Charlie Q it   doesn't have as many but there's some really  good ones and in the last mid Journey office   hours David did mention that they're working on  some sort of sraf Explorer no details on when or   what that will look like but it will be amazing  if they make an easy way to navigate to explore   these right on the site here's all my different  Wizards there's some amazing stuff in here [Music] [Music] another note on these is just like with  using images as style references you can   combine multiple srfs together just add  one then press space and add another now   it will merge those Styles the other  way is to click on the srf and it will   pop into the prompt bar I think these  two will be a cool combo here is from   those random numbers I typed then this  is the two I picked these are really cool you can also adjust how much of  each style is applied add a double   colon than the relative weight for  each so in this case the first style   will be twice as strong as the second  style now this is with those weights reversed personalization is the newest feature  they've rolled out and it is amazing usually when   you write a prompt the algorithms fill in all  the missing information with their own preferen   that gives mid Journey its own default  Aesthetics which is really the combined   preferences and biases of the entire Community  with personalization mid Journey will learn what   you like and fill in the blanks with your  own preferences so this will be completely   different for every person you can use it on  any prompt by adding d-p or you can go to the   parameters and switch it to on and it will use  it on every prompt without having to type it   each time here's what it looked like for the  prompt wizard with and without my personaliz my personal code is right here and I could share  that and anyone will be able to generate with my   personalization which leans a lot more dark and  surreal but I'd assume most people will prefer   their own codes that's kind of the idea there's  more to go over and I'll run a lot more prompts   with this but first to get one you need to rank  at least 200 images go to tasks then click rank   images and then it will have image pairs that  pop up so select which image you like better   choose based off of Aesthetics not how perfectly  it followed the prompt I don't even read the   prompts when I'm ranking these to go faster use  one to select the left image two to select the   Right image and three to skip sometimes you will  have completely different prompts and styles other   times it will be the same prompt with similar  images just subtle differences then if you're ever   just not sure which one to pick click three to  skip do this until you hit at least 200 I assume   doing more will help once it's all trained up  it will work anytime you hit d-p there's nothing   further you need to do I ran a few more simple  prompts with and without it for a comparison [Music] [Music] [Applause] [Music]   I love what mine does on these simpler prompts  I've been blown away at how much more I like them   actually here's on some of the other prompts  used throughout this [Music] video sometimes   when I'm going for something more specific it can  conflict with what I'm trying to get so I'll have   to to turn it off for certain types of images or  adjust how much personalization is applied which   you can do with the styliz parameter normally that  applies to how much of mid Journey's aesthetic   is applied when you switch to personalization it  controls how much of your personalized aesthetic   is applied I've been liking mine at pretty  much every level you can also combine this   with style references and other parameters I'll  use these two srfs again I'll wait the first one Higher I'll go with with raw make sure  personalize is on I'll add just a little   bit of chaos then an aspect ratio of 32  and I'll lower my personalization a bit   at stylize 50 and wow these turned out so  good that style combo fits the vibe of this perfectly there's two more parameters I want  to mention I didn't mention them alongside the   others because they have very specific uses and  aren't as common first is tile this generates   images that can be used as repeating tiles  to create seamless patterns this could be for   textures wallpapers Fabrics or even something  like a 3D scene how about multicolored Stones   pattern first I'll send it without tile now I'll  add Dash Das tile at the end and send that I'll   download one from each of these and take it into  a seamless pattern Checker this first one doesn't   work now I'll upload the tiled version and it's  perfectly seamless now there's also super tiling   which is something I learned from mid Library I'm  not sure if he's the original person that came up   with it but I'll do a quick demo of that I'll  copy this prompt to get a where's Waldo style   I'll change it up a little bit how about a vibrant  futuristic city all right this one looks cool I'll   test it in the pattern Checker and that looks  really repetitive the super tiling trick will   make it less repetitive open that image up and  use very region then select most of the image   but leave some space on each side now when  I send that it will replace that area but   remain seamless and now when I put these  all together there's a lot more variety I   essentially have a much larger and much more  detailed tile now so here's the before and after another unique use for the tile parameter  is something I came across from Nick St Pierre   you can use it to convert an image into a seamless  360° photo so this is really cool I had actually   done this before using generative fill to hide the  seam but I'd never thought to try it with the tile   parameter I'll do a prompt with a panoramic  view of a volcano an aspect ratio of 21 then   add the tile parameter actually I ran that again  to add some lava and this one looks awesome so I   will upscale that then I'll use this website  I found for viewing 360° photos just drag it   in and it's looking pretty solid so I'll switch  to the VR View and this looks amazing amazing I   can rotate all around and never find a seam so  I would probably still use the generative fill   method since a lot of the results I got turn out  a little weird when mid Journey tries to make the   top and bottom tilable too especially when I tried  going wider so I just really quick show how to do   it in generative fill so this is one without the  tile parameter so it would not work as is I'll   bring that into Photoshop go to filter other  offset then drag it to the side of bit that's   good now I'll select the rectangular Marquee and  drag that over the seam and use generative fill   no prompt needed this one's a little weird  since there's light and dark right next to   each other but yeah it'll still work  that's it export and bring it into the viewer there was a really amazing example in the  comments on Nick's post of all these people on   the Titanic they used magnic to upscale you would  never get this many perfectly sharp faces out of   mid journey magnific is amazing it's hard  for me to ever recommend because it is so   expensive but the results you can get from it are  unparalleled the stop parameter will finish a job   partway through the process to create blurrier or  less detailed results so I'll grab the seed from   that woman I've been using you type Das Das stop  space and a number between Zer and 100 the 100   is the default basically just finishing the full  job so so as we're watching this generate at 100   basically what the stop parameter does is just  stopping it at different parts of this process   so here's how that looks at each of the different  stop values it can be useful at 80 and 90 to get   less detailed and softer images V6 tends to add a  lot of detail so I use it to lower that sometimes   I have found very few other uses than that I've  tried it impr prompts where you want like a shaky   cam or motion blur you know like someone being  chased through the woods but really I can get   a good result just by prompting the other thing I  have used for is when I generate something that's   going to be a background that I know I'm going  to add blur to in Photoshop so in particular for   thumbnail backgrounds I'll use stop at 60 to 80  I'll still typically add some blur later but I'll   be able to get a better idea of what it's going  to look like right within mid Journey so that's   helpful um that's about all I've ever used it for  where I actually found it useful multi- prompting   only works well on V5 and earlier this is mostly  because V6 is so much better with understanding   language it's just not necessary about how  this works just in case you need it separate   Concepts using colon colon as a separator you can  assign relative importance to each concept using a   number right after the colons so using space ship  that gives you a space ship but space colon colon   ship gives you a ship in space if you give more  importance to space there will be more space in   the image or more importance to ship there will  be a bigger ship and these numbers are all based   on their values relative to each other so 2 to  1 is the same as 100 to 50 and again that just   does not work very well in V6 there's also  a parameter D- video this one does only work   in Discord the only thing I've mentioned so far  where that's the case all it does is make a short   video out of the image being generated it does  look pretty cool to use it in Discord you add--   video to the end of your prompt then once the job  is finished click add reaction then select the   envelope emoji and you will get a DM from the  mid Journey bot with a link to view the video at so just another parameter that's there if you   ever need it it's not generative text a  video or anything although mid journey   is actually working on a video model  they say they are just not close on it I'll use this prompt again for one  more example I can get rid of this I'll   use this srf it's a really interesting one I was  experimenting with for parameters I'll leave it   at 169 add a little bit of chaos switch to raw  and turn personalize on this is a really strong   srf so I'll lower the weight to 20 I'll also  lower my personalization to 50 and that should   be good I'll send it these are awesome that srf  gives such a unique Vibe I've ran it a few more times I think I like this one the most that is essentially all I know about  mid Journey packed into one video I hope   it's helpful all of the resources are linked  down in the description and of course make   sure to check out futurepedia to keep learning  about everything related to AI find the perfect   tool for any use case stay up to date with the top  100 tech companies developments in AI with the AI   Innovations tracker there's a curated database of  AI tutorials tons of additional learning resources   and all sorts of things being added to the site  regularly sign up for the newsletter to get AI   news tools and tutorials delivered straight to  your inbox And subscribe to get more videos like   this one on your homepage thank you so much  for watching I'll see you in the next one