Inpainting in Midjourney: A Deep Dive, Secret Commands & Tips for the Long Awaited Feature!

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
hi everyone so it's here one of the most long awaited and most requested mid-journey features of all time in painting has finally arrived today we're going to take a deep dive into how to use it to see exactly how powerful it is we're going to go over some of the limitations as well and some secrets that I've uncovered that I haven't seen anybody else covering but most importantly how much fun you can have with it okay let's Dive In first off if you're unfamiliar with in painting it refers to the technique of restoring damaged or deteriorated parts of images or video the concept has its roots in the restoration of artwork but today we're talking about its use in mid-journey so really what this is about is highlighting an area and being able to re-prompt within that area so yes that means that the multiple fingered people are a part of the past not to say that we won't generate them it's just a lot easier to get rid of those extra fingers and limbs as we can see via this image by proper prompter for the record I'm just happy that this was not a panoramic picture of a man named Peter riding a porpoise by proper prompter nailed it it was like the fourth take that was actually pretty good so the very region is the button that we use for in painting in mid-journey the very strong and the very subtle are buttons that we went over in the panning video that I did I'm guessing that very region or in painting is pretty much going to be your go-to from here on out that said don't count very strong and very subtle out because there are still uses when you're looking at changing the overall aspect of your image so with very region which I'm really just going to call in painting from here on out you have the ability to select certain areas of your canvas and have mid-journey reimagine it without any additional user input mid Journey will decide what to put there based on contextual clues in the image surrounding the area if you want a prompt for specific things you're going to want to make sure that you have remix mode turned on to access it go to your forward slash settings and make sure that remix mode is turned on so to show you how that works this is a very simple Pro prompt I guess that ends up being Steve aoki's dad sitting on a couch next to a lion you know as you do to get started all we have to do is hit this very region button and within a few moments mid-journey has reimagined the scene in just that area in this case uh generating up I guess Steve aoki's dad's third wife giving him seductive looks as he naps it out and I'm not judging the man he may have had a very long day as an additional note aim painting only works with upscaled images so via mid-journey documentation here is a look at the in painting module as you can see in this corner we have an undo last step we'll go over that in one minute free draw the rectangle tool in which you select your areas with the free draw tool is a little finicky we're going to go over that as well in a minute The impainting Prompt when you have the remix mode turned on cancel return to Discord and then you can resize or move the window up in this corner although it is locked to the Discord window so you won't really be able to move it around if you're saying full screen mode it kind of is what it is as a note everything that you see on the screen is available as a PDF over on gumroad it is totally free but donations are of course always appreciated also if you're a member of the patreon you've already got this PDF waiting for you as a quick FYI mid-journey does advise that in painting works best when you cover a region of 20 to 50 of your image although I've gotten much much smaller than that and had very good results so there might be a cya kind of situation there in painting works really well when you combine it with zooming out although oddly you cannot impaint after panning I'm not sure why that might just be a temporary glitch that gets fixed later on but as it currently stands if you pan you then are locked out of in painting so let's take a look at a use case here this is the back rooms which if you're not familiar with is kind of a surreal existential horror collaborative fiction experiment or thing let loose on the internet that's a big influence on the TV show Severance which is amazing so anyhow it's the back room Style by Gregory crudson with a weight of two Style by bread and wolfel uh with an aspect ratio of 16-9 uh wolfel and crudson are both photographers that's something that I've been experimenting around a lot with taking two photographers waiting one and then mashing them together you get some really interesting results that way kind of a hybrid of the two Styles so obviously because I didn't give any character details we ended up with the very typical mid-journey character with back to us but now within painting we can change that so by moving into our in painting module and just grabbing the rectangle tool I can just highlight her and then changing the prompt to a woman in a white shirt and a black skirt comma Sad Eyes by using sad eyes we've indicated to mid-journey that we want to see her face writing that gives us pretty much the exact same background as promised as this is in painting with our subject now turned to us with her sad eyes um I'm very impressed with the way that it handles lighting although it's not completely accurate here in terms of motivated light you know we've got our light source here it's kind of creating a Halo shape here but there's definitely another key light coming in from the front of her shooting down I mean this is how you know a set would be lit as well it wouldn't necessarily be completely motivated by that light in the background but she does match in terms of the overall lighting and composition within this Frame so just to give us a little more room to play with I took that image and zoomed it out 1.5 times to give us this image and now because our subject here looks like the victim in a horror movie I decided to flip the script via in painting so I wanted to keep the subject mostly the same except have her holding a red balloon because let's face it if you stumbled across this girl holding a red balloon in this room um you know she's gone from victim to probably the daughter of Jigsaw so in order to accomplish that we just take our rectangle tool and highlight this area leaving both her head and her other arm outside of the bounding box although here we want our balloon floating up so at that point we can move over to the lasso tool and paint upwards here the lasso tool is a little bit weird because it basically I'll show you let's finish this off here just give a little bit of extra stuff there there we go so um yeah you can see it's really finicky um so it kind of works as a circle and then you can counterclockwise it out um yeah like I said it's going to take some getting used to it's a pretty weird tool it definitely does not work the way that the Photoshop lasso tool nor does it work how the pen tool Works in Photoshop um once we have like this you can just Undo It by coming up here and zapping it out so um you may have a little bit of difficulty at first playing with this but remember that undo button is your friend so after executing that prompt we now have this image which I think works pretty well I would have preferred to have her uh holding the balloon a little more down the Rope um the blue it also looks a little over exaggeratedly big but for the most part this is what I was mostly looking for there was also this image which I thought was pretty good the heart balloon was kind of a nice touch making it a little extra creepy but I do want to point out that there were some mistakes as well this one's okay but in this image we have the return of the double Libs so that's something you're gonna have to keep an eye out for when you're in painting so I will point out that earlier in my experimentations when I put the bounding boss completely around her body we would end up in some weird poses uh like this one where she sort of like a balloon or in this one where she's kind of awkwardly holding it like a balloon for you but if do have to say I think when you compare our initial image to our final image it really showcases exactly how powerful in painting is particularly when you combine it with zooming in this we've managed to completely recontextualize our subject while keeping the background location pretty much the same impainting does work with niji as well this was cinematic action scene from a James Bond film aspect ratio 16 9 we got our bond-ish character here kind of doing the Leo walk away from an explosion so in painting that and just kind of giving it a bounding box here and writing in ninjas because you know if you want to make something cooler add ninjas which gives us this which truthfully maybe I should have just used a longer prompt but it does go to show that it does work um with a longer prompt maybe I wouldn't have ended up with this kind of emoji ninja over here I would say actually out of the four of them probably the fourth one probably looks the most in tune with what I'm looking for although it also kind of just looks like Emily the strange as an ninja I don't know I'll take it but he does go to show that impainting is a pretty powerful tool through it we can change ethnicities we can change emotions and we can add props let's take a look at those real quick long time viewers of the channel will recognize our old friend the man in the blue business suit walking down the street um so issuing a in paint on him by utilizing the lasso tool and just covering his face and issuing The Prompt Korean man we have now changed our men in the blue business suit to a Korean man in a blue business suit it's really impressive when you actually a b them um so just by hopping back and forth between the two and see what a good job in painting really does so this is Portrait photograph of Joan of Arc Style by Annie Leibowitz the famous portrait photographer large format camera 35 millimeter lens Studio lighting neutral background um yeah solid Joan of Arc moving into the in painting module selecting just this area of her face and putting in happy and running that gets us this image which you know isn't the brightest of smiles but again she's Joan of Arc you know things things didn't turn out so well for her overall I think it looks pretty good um definitely at least on par or probably better than photoshop's liquify feature where it makes people smile as well that one's always kind of looked a little off to me and finally returning to niji uh we had this image which was cinematic still Mad Max raising a pistol an aspect ratio of 16-9 um just highlighting this area and changing that to holding Katana gave us this in-painted image which um he is now holding Katana now there are going to be some times when in painting is just going to flat out ignore you but there are some tricks to get it to do what you want for example returning to a character that we've used here many times in the past are cyberpunk woman with white hair walking in the snow I wanted to put a white wolf next to her but no matter what I did the wolf would not appear here's another example where I tried still no wolf zooming out on the image to give us more space for for a wolf still no wolf But ultimately the solution comes via the slider method now I'm pretty sure I covered the slider method in a previous video where I went over waiting but basically the way that it works is that you want to issue a colon colon and then repeat part of your phrase now traditionally you would add a weight in to the end of your prompt but there is no waiting in image prompting so the way that it works here is by taking the same prompt and adding in white wolf walking next to her colon colon wolf City alley cold snowing it basically ends up splitting the prompt Now by no means am I saying it's perfect and that ended up being a lucky role after a number of these weird hybrid um are cyberpunk woman turns into kind of a wolf hybrid um then there was this one where it kind of looks like she's sort of riding the wolf then there was this one which was accurate it just wasn't composed very well so I'm just saying it'll probably take you a number of roles to get to what you're looking for but ultimately it does work and although there's some weird kind of smudging that happens in the background details what you could probably do is just stack the original image with the wolf image into an image editor and then kind of erase the smudgier parts or blend them together so although we can't use image prompting or waiting within in painting there are commands that we can issue a vme journey documentation the following commands are available chaos fast image waiting no stylized relax Style version video and weird now that's not actually entirely accurate again image waiting doesn't work because image prompting doesn't work so it does lead me to suspect that perhaps that will be coming in a future update and that would be super cool additionally I can tell you that I have tried version and that does not work as well the commands that I have tested that do work are chaos stylize style and weird as always be careful with weird because weird does get weird for example here is Cinematic still wide angle filmed by Quentin Tarantino a man sitting in a Los Angeles Diner drinking coffee black suit 1960s aesthetic an aspect ratio of 2-1 and painting to the right side of him and adding in a clown with a chaos of 100 gives us this and then to the left of him I added in an alien with a weird of 50. uh one thing that I'd like to point out though at how good in painting is if you look at the version without the alien you can see that the reflection of the palm tree is actually in the table and then when the alien appears not only has he ordered danishes by the way but the palm tree has vanished and so has the reflection in the table the post reflection remains in both though so yeah that's pretty remarkable so let me know what you guys think about the new impainting feature and if there's any tips or tricks that you've picked up along the way I thank you for watching my name is Tim
Info
Channel: Theoretically Media
Views: 20,347
Rating: undefined out of 5
Keywords: midjourney tutorial, midjourney inpainting, midjourney prompts tutorial, midjourney ai tutorial, inpainting, midjourney remix, midjourney niji mode, midjourney niji, midjourney ai, ai art, midjourney prompts, midjourney control pose
Id: 7uPdvqpeEtU
Channel Id: undefined
Length: 13min 58sec (838 seconds)
Published: Tue Aug 22 2023
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.