The Good, the Bad & the Ugly of Midjourney INPAINTING

medrini finally released their interpretation of in painting just over a week ago obviously this got me excited because in painting has been one of those features that we've all been desperately waiting for however I didn't want to rush out a video without having sufficient time to properly test it so that's why this video may seem a little bit late to the party however as usual as the feature may be it's still far from perfect and if you're struggling to make certain things work for you then don't worry you're not alone so today I'll walk you through the Good the Bad and the Ugly of modernity's new very region feature all right so today I'm going to walk you through a number of different use cases and things you can do with majority's new in painting feature which within their ecosystem is called very region now before we can actually do anything you first need to make sure that you have activated remix mode in order to do that what you do is you enter slash settings and then you just make sure that this remix mode button is activated with regards to the high variation and low variation mode buttons I must admit that I'm not entirely sure yet to what extent they actually make a difference in this particular case when we use very region that is something that we'll have to discover as we go along and in the next couple weeks but for the purpose of this video I just decided to stick with low variation mode all right so let's start off with our very first prompt so for our first use case we're going to look at adding and removing objects from an image so let me start off with the following prompt which is Cinematic photography astronaut exploring vast deserted landscape on distant planet Venus plus Mars and then I add some parameters let's wait for these images to generate all right so let me just move myself over here to the side and let's have a look at these images and sure enough it's exactly what we would expect from this sort of prompt we've got four really nice vast Landscapes on some distant planet and we've got our astronaut in each one of these images now I think what we're going to do in this particular case we're going to start working off of the bottom left one simply because that's where the astronaut is most dominant and let me just quickly upscale that image and once we've done that we can now start working on this image using very region now what I'd like to do is I'd like to remove our astronaut from the image now I know it doesn't really make sense to First prompt an image that includes the astronaut just to remove him afterwards but the point here is just to illustrate what you can do so what we would do is we would hit the very region button it opens up this nice little window which is quite neat actually because I'm curious to see what majority does with this because they are working on a web UI but it seems that this might actually be a good alternative for all of those who don't necessarily want to work within the web UI anyway so what we're going to do next is we're going to pick one of these two buttons so you've got the rectangle tool or the lasso tool in this particular case I think the rectangle tool will be fine so what I'll do is I'll just select like this area here and I'm going to change my prompt to cinematic photography deserted landscape on distant planet Venus plus Mars now note that I've simply left out a whole bunch of stuff from the original prompt I didn't really have to change all that much I do like to keep the main elements that Define my style just to make sure that my journey and myself are more or less on the same page and once we've done this we simply hit this button here and wait for it to process all right and this is actually an excellent example of just how imperfect this feature still is remember we selected the entire astronaut we entered a prompt and that did not make any reference to an astronaut or any other character and yet even though my journey removed our original astronaut it re-inserted some type of character now this is one of those things that most people on social media will not show you they will mostly just cherry pick the best ones and show you what's possible but rarely will they show you where majority falls short now this isn't a big issue because what we'll just do is we're just going to upskill one of these again we'll take the bottom left one again and once it's upscaled we'll simply hit very region one more time and this time we'll select this character down here and actually that might be not enough area and we'll keep the exact same prompt and hit it one more time now this time around I would expect that character to disappear but who knows we'll see all right excellent so now we have these four images that look more or less the same where you've removed the character we don't have an astronaut in the scene anymore but now let's say we want to reintroduce some new objects I want to insert some structures so what I'm going to do is I'm going to upscale one of these images and then once it's upscaled I'll be able to hit the very region button and now let's say I want to place something around this area over here or actually maybe maybe that's not the best place to do it I'm going to put a little bit more in the to the distance over here and make it a little bit bigger instead and what I'm going to do is I'm going to use cinematic photography monolithic city-like structures in deserted landscape on distant planet Venus plus modific again we haven't changed much the only thing I've really adjusted is that I've added this monolithic City like structures element into the prompt I'll hit the submit button and we'll wait for these to generate all right cool so let's have a real quick look at these and what you can see is we've successfully inserted these structures I mean we're not really sure what they are but we've successfully inserted them into our scene and this is exactly what the new very region feature is actually quite good at you can remove elements out of a larger scene put in new elements in specific locations it won't necessarily always work perfectly but you can actually construct quite nice scenes this way now one area where I find this to be a little bit more challenging is once you have a scene that is a little bit more condensed where you've got a lot more going on in a much smaller space so to demonstrate what I mean let me show you this prompt interior architecture photo of living room of luxury condominium in the Swiss Alps earthy tones warm cozy and let's have a look at what we get from these okay so these are really nice and they're exactly the type of images that you would expect from this kind of a prompt um The Journey just gives us these really nice interior designs now I'm going to work off on the bottom left one again simply because I want to demonstrate something which I find a little bit frustrating you can see that there's this little pillow or mini sofa like object here in the bottom right corner I'm not exactly sure what they're called but anyway what we're going to do is we're going to upscale number three and what I want to do is I want to remove this object and this is something that similar to what I showed you in the earlier example with the deserted landscape this becomes very difficult to do with these interior designs or not very difficult but a little bit annoying so for example I would hit the very region button I would then perhaps use the lasso tool this time I would select this area where we have the pillow or the e-mini sofa and what I would enter is something like carpet living room now this is just an example of what you could do but I just want to show you what happens if I do this sort of thing hey sorry to interrupt you right smack in the middle but I just wanted to share something with you YouTube is an absolutely amazing way to discover and learn new skills just like the video that you're watching right here however as you discover these videos and tutorials you're never quite sure when they were recorded whether they're still relevant and how they fit into the bigger picture of learning to use and control midtermee so if you're looking to get a comprehensive training in the foundations of prompting in my journey and I highly recommend that you check out my video course at mastersofmanjourney.com which gives you a comprehensive education in learning how to control my journey and how to prompt with intent we can submit all right and actually in this particular case it turned out pretty okay so we have two images where the pillow has disappeared the other two images it hasn't quite worked yet and very often what I've experienced is that none of the images will actually work simply because the it I can't really explain what it is but it seems that the more closely positioned the different objects are the more difficult it is to remove them and my suspicion is that this is because majority is actually not only looking at the area that we've selected but in fact I think it's actually taking into account a lot of the context of the entire image and to an extent that might be a lot more that we would like it to do so in this case it worked out great but let's say I did not well what I would then do is I would simply upscale well one of the images that was the closest to what I was trying to get to and I would then hit the very region button one more time and then very often what I've noticed is if you select a wider area then things will work out a little bit better so if you you don't follow the Contours of the object too closely but instead you actually select a much larger area around it now this obviously won't always work because there might be other objects in the way but in this case it you can do it and so I would enter something like empty living room floor and then maybe use a negative prompt here I'm honestly not sure whether these negative problems work within this feature but you know we might as well give it a try I hit submit and this is precisely what I meant we have another look at two of these worked out fine whereas in the others it did exactly the opposite of what we wanted to do now of course we got two which are fine so I should be happy with this but it just goes to show you that we don't actually have as much control over this feature yet as we would probably like and there's I'm sure there's a lot of fine-tuning that the majority pimp still needs to do all right so now let's have a look at something else that you can do with this new feature all right so I'm going to start off with a new prompt which is just slightly different from what we used before interior architecture photo of living room of luxury condominium in Santorini Greece earthy tones warm cozy summer bright sunlight let's generate these and then we'll see which one we'll pick all right so before I actually open these I gotta say I'm in turbo mode right now and for some reason the journey just pushed me into fast mode I was just a bit surprised that with this was generating really really slowly compared to what I'm used to and this is a big question in my career I don't know why this is happening um maybe some of you guys know and if you do please put into the comments but this is a bit irritating anyway let's have a look at the images and these look beautiful if you've ever been to santoini then you know what that this is actually pretty close if you have not then you should go at least once in your lifetime but what I really want to show you now is I want to show you how we're going to replace Scenic elements within an image so we are in santoini and the interior architecture is exactly what you would expect from that type of location but what I want to do is I want to change the outside scenery to something that might look a bit awkward but I just want to show you what this feature can do so I'm going to pick one with lots of Windows which is the bottom left one again it's interesting number three is the number of the day it seems and we'll wait for this top skill and once it does we'll open the very region feature and what we're going to do now is I'm going to see if I can zoom in a little bit Yeah and then I will select all of this in the entire area that shows anything that's on the outside and I'm going to try to hit the Contours this time because otherwise it'll produce some well relatively strange results and actually I messed up here let me add this as well then there's a window over here which I'll also cover and we'll do this one as well let me zoom out and what I'll do is I'm going to use the following prompt and New York City skyline during sunset now I know this interior design does not fit to New York City but I just want to prove a point I'll hit submit and we'll wait for these to show us some nice new images all right let's open these up and have a quick look and as you can see while it may not work well in every single one of these because some of these images look a bit strange to be honest it does seem to have put New York City into the Horizon of our Skyline and it even filled in the other windows as well and I know this looks out of place but it just goes to show you that if you want to do something strange it does work now the one thing that bothers me about this though is that it does have its limitations again so let's say I were to redo this one more time so I'll hit the Embry region button again and this is very important if you work off of the same image again but you want to select something different it's very important that you actually hit this undo but in first until all of these selected areas disappear because otherwise you're going to start processing on different areas at the same time I made this mistake a couple times before and then I had to redo the same prompt again and again in this case we're going to work off at the same area and actually what I might do is I might expand this a little bit more just because it looked a bit weird just a second ago and we'll do is we'll pretend that the room ends over here we'll just select all of this and we'll add this space as well but what I really want to show you is what happens when I enter a prompt like this thick snowflakes falling Swiss Alps and what I'm hoping to see is a panorama or like landscape on the Swiss Alps and I want it to be snowing one of the things that I find frustrating about this is that the way the impacting feature works is it it doesn't it's not as faithful to our prompts as if you were prompting an image directly and this is important because it means that you do not have as much control as you might think and I again I think it's taking into account a lot of the context of the Total Image including some of the lighting the the furniture and everything and it just then makes a decision to not give you what you want so let me show you this hit submit okay and let's have a closer look and surprisingly enough this time I actually got some slow flakes here on the left they do look a bit weird but on all my previous attempts it just didn't want to work but then again if we look at the other three it's like bright sunlight and like sunshine there's not a single snowflake in sight and even the one in the bottom left that has snowflakes it doesn't really make sense because it has like it's blue skies and everything so again I think the journey is looking at the overall image and based on what it sees it then decides on how to interpret the prompt that you enter for the in painting and again it's I mean if you know this if you understand this then that's great because then you can adapt to this but it also means that you're being restricted in what you can do and I I find that a little bit disappointing to be honest anyway so that's enough for interior design let's move on to the next examples now one of the things that you can use major Nissan painting for now is a certain level of face swapping it's not 100 the same thing as face swapping with other tools but it comes pretty close and let me show you with an image that we'll start off with street photography close-up portrait of a firefighter in full uniform let's quickly open these up and as you can see these are really great images but what I want to do now is let's say I've decided that I really like the face of the character on the top right but I want to use the image of the character on the bottom right now they don't look exactly the same so what we're going to do is I'm going to upscale both of them skill u3 2 and u4 and what we'll then do is we will copy the link I really gotta say my journey is super slow today and I've it's gotten worse over time I don't really know why but anyway so I'm going to copy the link of the face that I want to use and by the way if you're not familiar with this toolbar down here this is my prompt about Chrome extension which makes prompting so much easier if you want to grab it yourself you're going to have to use this within the Chrome browser or a chromium-based browser this this does not work directly within the Discord app so you're gonna have to use Discord within the browser and you can grab this at www.pumpthelot.com extension anyway so I've copied this the image URL of this face and I'm going to go down here to the one where I want to face swap it into that looks like very region and what I'll then do is I'm going to use the lasso tool to well actually I don't necessarily need to use lasso tool since this is actually pretty much square so we can just do this with a square selection tool and then we will paste this down here this is the URL and then I'll keep the rest of the prompt the same to be honest because I don't necessarily want to change anything else so I'll hit submit and if we open these up and then also show the original images next to these then I think you'll notice that we've managed to inject most of the facial features of the well not the original image but the one that we want to use and we still have the overall composition and the the shot of the other image now again this is not 100 perfect and it works best when the shape of both heads is similar as soon as you have a face that has a very different shape as the other or maybe even a long beard then it becomes a little bit more tricky because it will take the facial features but not necessarily the shape of the head and then it just looks a bit weird but anyway this is something that definitely works pretty well and I think it might be good enough for many many different use cases now I want to show you another thing that we can do with faces okay so let's say we're happy with either one of these faces I'm going to select maybe the top left one and the only thing I want to change though is the facial expression the emotion this guy for example has a face that's perhaps a little bit too nice and what I would rather do is I'd like him to look a bit more Fierce and angry so what I'll do is I'll hit very region I will once again this time I might actually just use the lasso tool so let's select only the face you don't necessarily need to do this I'm just experimenting here on the fly to be honest and what we're going to do instead now is we're going to enter a few extra words close up of and oops sorry close-up portrait of angry firefighter and at the end I'll end upset frustrated this will still use the original reference image I'll hit submit and let's have a look at how this changes the facial expression of this character all right let me show you these real quick and as you can see it's taking still most of the facial features of the face swap image and the only thing it's done now is it's actually given the character an emotion the man is very angry he's frustrated he's upset his mouth is open so this is something that is really great and gives you so much flexibility in terms of changing the faces of your characters in particular scenes and just making them look a bit different and it's going to give you much smoother and much more polished images than anything that you could have done before now let me show you another thing that you can do with characters so what I'm going to do this time is I'm going to enter a new wavefront street photography full body shot our fashion model walking through the streets of Amsterdam Autumn season let's generate these all right let's open them up and sure enough they look great this is the kind of image that you would have expected but perhaps we want to change aspects of this character now I'm going to select the top right one U2 and I'm going to do a few things one after another so first of all what I want to do is I want to change the woman's hair color so I'll hit the very region button I will now use the lasso tool just to make it as specific as possible I will select all of the surroundings of her head and her face actually I don't know whether I need to pick I'm just gonna since I don't care about the face too much I'm just going to select that one as well and I'm just going to enter brunette woman and sure enough this is still the same shot same composition but we now have a brunette in this image instead blonde now let's say we're not fans of our outfit either and we wanted to change this well we could do that too we could potentially give her a new set of pants maybe even a new sort of shirt maybe not a shirt maybe a turtleneck instead and you can slowly bit by bit change the entire outfit of this character so I'll hit very region one more time and this time around I'm going to select only her pants and I'll bear in mind we're going to mess up the hands here unfortunately but that's oops sorry I messed up here we're going to mess up the hands but that's fine because we can fix that afterwards as well that's also another huge benefit of this new feature and what I'm going to do is I'm going to choose aubergine Chinos let's see what we'll get from this prompt okay so unfortunately the color didn't really get through in this attempt except for the bottom right one maybe but what it did do it fixed most of the hands which is excellent already and I want to show you one more bit here because I really want to do the turtleneck so I'll upskill u4 and I'll hit very region one more time just to make our last change the lasso tool and then I'll just select this little section here up until just enter just until your neck select all of this and then what we'll say is we're going to take ah an ocher I mean these colors really don't work together well I mean I'm just picking something to you know make a point but an ocher turtleneck let's submit that and have a look at what we get interesting so again we didn't quite get the color which is which is interesting honestly because in most of my early attempts it went really well but this particular case just kind of it didn't mess up we still have a nice turtleneck and it fits to the overall outfit I guess but if you wanted to get the right color you may have to repeat this process multiple times until you actually get to that color so in that sense it's actually similar to using remix in you know the classic scaffolding remix where you would have to reinforce over and over again until you get to your result and it looks like this is something you may have to do with these images as well so this just shows you how you can actually mix and match an entire outfit and I think there's going to be a huge game changer especially for people who are working in the fashion industry or you have an e-commerce store and you want to change like outfits of people this is gonna just blow people's minds all right so I now want to show you another not so great example so in this example I'm going to show you what happens when you try to replace a complete character and just switch them out with somebody else so I'm going to enter the prompt street photography to police detectives observing a crime scene we'll generate these images and again these are images that we're used to they look really really great but maybe I want to replace one of these characters with another character and not just swap out the face but actually we want to replace like the gender so this is a guy these are two guys in both cases I think maybe in the top left that might be a woman I'm not entirely sure but what I'm going to try to do is I'm going to try to use the bottom left one here and replace the detective on the left with a female detective so I'll hit the very region button and to make things a bit easier I'll just select an entire section here and I will replace The Prompt with female police detective observing a crime scene I'll submit this and what I'm kind of expecting to happen is I might get a woman but then probably she'll look really just full not distorted but just out of place or the character will just disappear altogether and yep that's kind of what we're getting here as we can see it's gonna take a little moment to finish all right and this is one of those frustrations that I have with this feature now I must admit that in two of these it actually worked quite well so I'm not really sure what's going on today because it seems that my journey might be fixing things in the background but so the one in the bottom right actually looks pretty good the one in the top left looks pretty good as well but the other two now like why like why how does how does this character just disappear and in my previous tests I would literally have four Images where the character would just completely disappear I would like there would be nothing or there might be a woman but she would look really weird like way too big or way too small or just like with a cropped body and so that makes it kind of difficult to work with this if you don't really know what you're going to get it's like it's way more random than what you would usually have with metrony but anyway so this is to show you what happens when you use this on an entire character now before we finish off this video I want to show you some really really horrible examples of what happened in my earlier tests and for this to work I'm going to have to open up a an image that I saved and I'm going to show the prompt on screen just for you to see what I prompted but we got this image of a crime scene in New York City and what I wanted to do is I wanted to place a female detective once again somewhere within the scene as part of the greater scene so what I did is I selected the I selected an area in the sort of left section of the image like a rectangle and then I prompted the I used the following prompt film still 40 year old female detective with light brown long hair that's all I entered okay and what I got was these images which are just creepy like what is this why I prompted a woman a female detective and instead it placed these what is this even these are four creepy dudes with long hair their bodies are chopped off and kind of like cemented into the ground like and I don't even know why does this guy have like these neon lights over his body it's so this was one of my earliest attempts with Barry region and it I was a bit shocked because it felt like nothing is working so what I did is I made another attempt and then I used the following prompt film's still full body image a 40 year old female detective with light brown long hair standing in Street looking at crime scene in New York City and what I got from that was the following image so again one of these actually looks pretty fine the one on the top left but she's kind of not looking at the crime scene she's looking at me but okay I'll let that pass the bottom left one again is a cropped woman I don't know where that lighting is coming from either because her face is clearly lit up um and then in the other two minutes there's dudes there's guys like like how difficult is it to deter German whether I want to see a woman or a man so again I think what's happening here is that madrini is using a lot more context from the overall image to the point where it's starting to mess around with what you have actually asked it to do and I understand there's a fine line between you need to find a balance and how to interpret what the person is prompting and you know how to fit into the overall image but it's it feels like this feature still needs a lot of work that being said based on what I've shown you today I think there is lots to be happy about because we're now finally able to fix small problems and make adjustments here and there add stuff to a degree which just wasn't possible before and even the face swapping I know it's not perfect but it is a big step forward and may allow you to create much more polished face swap images than what's possible before all right so that's it for today this is my quick recap of everything that I've tested so far with very region and I think there will be much much more to look forward to in the next few weeks remember to check out the description for my majority course as well as a whole bunch of other free stuff and uh yeah I'll see you next week and take care keep on learning thank you [Music] [Music]

Channel: Tokenized AI by Christian Heidorn

Views: 6,382

Keywords: midjourney, midjourney v5, midjourney version 5, midjourney tutorial, midjourney prompts, midjourney workflow, midjourney learn, midjourney 5.2, midjourney 5.3, midjourney inpainting, midjourney vary region

Id: bRYFObeyzR0

Length: 25min 37sec (1537 seconds)

Published: Sat Sep 02 2023