Underused Midjourney v5 Prompt Commands :: How to use Text Weight and Image Weight

hey everyone today I wanted to expand a little bit on two pretty powerful techniques you have available to you in your mid-journey prompting Arsenal namely Image Weight and text weight I think they're underutilized because there's some confusion as to how they work so in this video I'm going to talk about what weights are how they work and how you can combine text and image ways together to really control your output but we're also going to go over what some of its limitations are and how we can get around those as well all right let's Dive In before we dive too deeply into weights I think it might be a good idea to briefly explain how prompts work in mid-journey every time you prompt something mid-journey scans your prompt and looks for keywords and then assigns tokens to those keywords those tokens are then used to assemble your image based on what mid-journey knows in its database I I don't know what the exact number is but I've heard that you're assigned about 75 tokens per prompt I'm sure that number has gone up as mid Journey's team hones in on the language model so taking that hypothetical 75 the mid Journey bot scans across your prompt looks for keywords and then divvies up that 75 across but here's the thing it doesn't do it equally it definitely places more emphasis on the words at the start of your prompt than towards the end for example with the prompt a cupcake a cup of coffee a napkin a wooden table we have this image actually that looks a little more like a muffin in my opinion but that's kind of splitting hairs but emphasis definitely placed more on the baked good than the cup of coffee however if we swap the order of that around to a cup of coffee and napkin a wooden table a cupcake we end up with this image which definitely favors the coffee cup over the cupcake now to be fair it did take a couple of rolls to get there I think that happens a lot with you know mid-journey YouTube tutorials is that people tend to cherry pick like these example that worked so I do just want to illustrate it wasn't every single time I got like these weird ones where it forgot the coffee cup or you know the napkin sometimes it takes a little work to get what you're looking for but I think waiting is an important tool in getting there text weights are our way of letting mid-journey know to add more tokens to any particular keyword that we choose to weight something all you need to do is in lieu of a comma use colon colon one and really any other number that you want or even decimal points if you don't add a number after the colons mid-journey just considers that a one so if you were to take your first keyword and do colon colon with no number it's a one and then on your second keyword if you do colon colon two it will then treat the second word twice as important as the first one the same logic applies if you do colon colon 50 for your first keyword and then colon colon 100 for your second keyword it would still be a two to one ratio for your own sanity I would recommend numbers between one and ten but you know if you've got a head for math go ahead and make the median like 1 325 it you know it'll still work it's just you'll have to do a lot more calculations one important note on the format of waiting and it's something that I've seen in the community feed and in a couple of other posts prompted is that there is no space between the keyword and the colon colon but there is a space after the colon colon to your weighted number and then following that there is no comma to the next keyword when that happens you might still get an image off and think you know oh mid Journey's not listening to me but what actually happened is that it was incorrectly formatted so mid Journey just ignored it so as a very basic but kind of interesting example let's just take the word cupcake right cupcake aspect ratio 16 9. they look delicious I kind of like the third one I think the third one looks the tastiest which one do you guys prefer which one is the tastiest looking to you so if we separate the word with a comma into cup comma cake we get this which is interesting right it's not necessarily what you think that we would get if you just read the prompt you would think that we would get a cup and a cake next to it but instead mid Journey creates this what ends up happening here is that it looks at the comma as a colon colon so it has the cut part of it and the Cake part of it and then once you combine it into one image so hence a cake in a cup and to be fair the easiest way to attain that result would just be to use natural language but I think this is a good example and illustrates pretty well how image waiting works so for example if we do cup colon colon 0.5 with cake weighted at two we end up with this image which favors the cake and b cup is now in the background and we kind of it's not a full full cake but it's you know still a cake again I think the easiest thing in this situation is just to use natural language like this is a cup of coffee next to a cake and we get the exact results that we want but where I think waiting really shines is when you have a longer prompt with multiple elements and you want to play around with the compositional balance for example this is an old Samurai rough beard holding a katana in a mystical Forest looks great great pretty much exactly what we asked for so let's play around with some weights and see what happens SO waiting Katana at four and mystical Forest at two gets us this uh it's interesting that the first image shows that the composition has completely changed and is very much favoring the katana um the other images yeah so-so-ish but really that first image is kind of the winner for another example I ended up cranking mystical Forest up to four and rough beard up to two leaving the samurai in the katana at one and we got these results and you can see that it's treating that mystical forest with more importance by making everything a wide angle to show more of the scenery that third example I'm not sure exactly I guess it's really focusing on the heavy beard maybe as an alternate but for some reason it just kind of gave me a like a hipster photographer um that definitely does not look like an old Samurai but as a starting place for a prompt it's actually not that bad and I think that by adding to the prompt you could really fine tune it to get something that you were looking for and very quickly before we move into the next section on image waiting I just wanted to ask if you haven't had an opportunity to like And subscribe if you could please do so I really am trying to grow the channel it would be appreciate it if you did all right let's dive in so I'll know how to use reference images in mid-journey you upload an image to the Discord server you take that URL and then put it at the front of your prompt and it will act as a reference image but I don't often see people using image waiting which you can do in mid-journey uh via dash dash IW you can score an image weight between 0.5 and 2 which tells mid-journey how much you rely on the reference image to its final output I do think it's important to note that image referencing is not the same as stable diffusion's posed image rather the output is something that's inspired by your reference image if you want a direct one to one you might want to check out the video that I did on Leonardo's post image the link is below but that is not to say that you can't get pretty close in mid-journ so I took this image of Scarlett Johansson as the Black Widow doing the superhero pose um and gave it a prompt a female assassin red hair black Outfit Black Widow in the style of J Lee uh Jaylee is a comic book artist who actually has done some covers for the Black Widow Within an image weight of 0.5 and these were the results which I wouldn't necessarily say is very J Lee in style but you know it definitely got the whole Black Widow thing so let's try cranking the image weight all the way up and see what we get and here we get something much closer to the pose although it did lose all of the J Lee isms of it but just to show how you have to play around with things a little bit I took Jay Lee and gave him a weight of 10 and these were the results which pretty much still look very photographic so I then took the image weight down so again same prompt with an image weight of one and now we get this which is starting to look a little more illustrating not completely in that J Lee style but still like it definitely is now incorporating elements of the illustrator style that I'm looking for so let's move on to a different example where I can show you some ways of maximizing your image results and some tricky things that mid Journey does when you negative prompt so this was for our project I did a little while back it was a fictional documentary about the making of a Dark Tower movie directed in 1980 by Steven Spielberg and starring Clint Eastwood you don't really need to know much about the Dark Tower or anything we're just going to pretty much stick to the images here but if you are a fan of Stephen King's The Dark Tower you know you can check out that video it's on the channel so a prompt that I used in that project was Action Shot young Clint Eastwood as the Gunslinger firing his gun in an underground mining tunnel old west Steven Spielberg Western film horror film full body outstanding cinematography Style by 1980 70 millimeter I actually used V4 for that project and getting Clint Eastwood's face was a real hassle back then so but apparently it comes in pretty okay now the images are okay they're not as dynamic as I want them to be so let's use an image reference to see if we can kick that up a little bit so using a reference of Harrison Ford as Indiana Jones it kind of works I mean he's got that hat it's not quite a it's a fedora it's not a cowboy hat but let's see how that works with our prompt using an image weight of 2 gets us this which has the Dynamics that I'm looking for but you know it's very Harrison Ford so let's try dialing it back and see what we get dialing back to a 0.5 gets us this which is definitely more Clint Eastwood now you can see that in that second image and I guess kind of in that fourth image but he's got that weird glow on his gun there that's sort of strange so since my journey was having a problem with getting that Dynamic pose of Harrison Ford and the face of Clint Eastwood together uh I went back to the old trick of photobashing Clint Eastwood's face onto a Harrison Ford body actually looking at those fingers it looks like I photobashed um Clint Eastwood's face onto a mid Journey V4 output but that ultimately got us here which I think like probably the third image probably be my go-to maybe yeah probably the third image so yeah we got there so now let's experiment with some negative prompts and see what we can do a negative prompt is just adding a word in colon colon and then a negative number and that should hypothetically remove things or tell mid Journey you don't want to see those things um so we're going to start off on easy mode and just go with Clint Eastwood's uh belt and his holster so hilariously what mid Journey did when I prompted uh negatively on belt and Holster it just moved it into a you know mid to close-up shot it's pretty tricky mid Journey pretty tricky technically it did give me what I asked for so instead of continuing on that path I just decided to torture myself and try to remove the Hat um that I knew was going to be a little bit difficult given that the image reference has a hat and an end no matter how much you text prompt I think that mid journey is just going to want to put that hat in and indeed after numerous attempts that hat just would not come off so we moved into going back to photobashing I found a random image of Clint Eastwood from Hingham High did my fairly terrible photobash with the Harrison Ford image I ended up with this image which doesn't look too bad I think that that's something that I could work with and possibly take in as another image reference and do the whole process over again but again this was just as an experiment to see if I could get the hat off which via straight text prompting no but you know via photo bashing yes so hopefully this was helpful to you and gave you some ideas for some future prompts if you have any questions comments or suggestions please let me know in the comments below I always love hearing from you guys and as always my name is Tim and I thank you for watching [Music] foreign

Channel: Theoretically Media

Views: 6,302

Keywords: midjourney prompts, Midjourney, midjourney ai, midjourney tutorial, midjourney v5, midjourney weight prompt, midjourney weight image, midjourney weights, ai art, midjourney commands, midjourney ai tutorial

Id: MWVELgLlGgk

Length: 11min 38sec (698 seconds)

Published: Wed Mar 29 2023