Lessons

Lesson Info

Generative AI in Adobe Photoshop

Adobe has recently been adding features that utilize artificial intelligence in Photoshop. And it's been great as far as being able to get selections to be made without having to do a lot of manual work. And some of our retouching tools now are much smarter than they used to be. But a whole another category of that would be known as generative A I and generative A I is where it doesn't need to see an original picture to begin with. You can just create something by typing in text. The text is usually known as a prompt and it will simply generate whatever you describe and this can be used in many different ways. So we're gonna take a look at it. Now, when we do, I'm actually gonna use the beta version of Photoshop because what happens is Adobe puts new features in the beta version and then not all that long after that, they'll move them over to the main version of Photoshop. And so the beta version is something you can download. So if there's a feature in here that I mentioned, that's no...

t in the normal version, you could download that beta to get it or it might be rather, I wouldn't be surprised if by the time you watch this video, it might already be in the main version because it usually doesn't take them all that much time uh to get it transferred over. But I mentioned that because if you happen to watch this class, like the day it comes out, then you probably have to download the beta to get to some of these features. So let me show you that download the beta and then we'll look at generative A I so to download the beta. The first thing I need to do is come up to the creative cloud icon. It's just usually on my menu bar up here and I'll click on it. If you don't have it up there, you could look in your hard drive, it should be the same place where all your other applications are. And in here, I'm gonna choose the choice on the left called apps. And then at the top, there are various categories. And if I make this bigger, I might be able to see more of them or just there's an arrow here that allow you to see additional ones and you want to go to this one that's called beta. And if you choose beta right here, it will be Photoshop beta. If you don't have it installed yet, it will have an install button. And if you already have it installed, you can hit the open button if you want to bring it to the front. Now, the type of generative fill we've used in the past and other sessions was if you have something in your picture that you didn't want, then you could come over here and make a selection around it, including all of that particular object and then you could tell it to generate an image and you wouldn't need to supply it with a description. If you just want to retouch something out, you just need a relatively accurate selection so that none of the information that you want to remove is excluded from the selection. The selection would be best if it goes at least a few pixels outside. And also it would be useful if you included all shadows that the uh object you're trying to remove. Uh is it, you know, it has and then there's a couple of different ways to get to generative fill. One of which is to go to the edit menu and up there you find generative fill. Now, I'm just gonna leave this empty and hit generate and we'll see what it comes up with for this particular area. Now, when this happens the first time, I'm gonna just let this progress bar go. But in some of the future times I apply it, I'll fast forward and it'll look like it happens instantaneously just so that I don't waste your time just staring at progress bars after applying generative film. If you come up to the properties panel, you'll find that there are three different versions that are generated and so I can switch between them. This one has some odd text like stuff which I don't think is appropriate. So I'll try the second one which looks really good and then the third one and I think either the second or the third would work great. I'm gonna go for the third one because I think it has more of the distressed look that was originally here. Now, if you didn't like what it ended up doing, you could always just hit the generate button again and you get three more variations of this. And if you want to reduce your file size a little bit, there's no need to keep the ones that you didn't like, especially this one that had the odd text in it. So when you hover over this, you'll find a trash can in the lower, right? And you can click on that to delete the ones you might not want, but these are taken up space in your document. So if you know, you're not, not gonna use the other one might as well trash it to save a little bit of space. Whenever you do generative fill, it'll show up in a special layer. And if you look in the lower right corner of that layer's thumbnail, you'll see these two little kind of crossy things that I think they're supposed to look like magic Pixie dust or something. And that's an indication that this is a generative fill layer, but we've already talked about using it for that particular purpose. Just wanted to remind you and to let you know it's related to what we're talking about now. But now let's explore the other features that are available when using generative fill. I'm gonna switch over to the crop tool and that's another spot where you'll find a related feature. Up here in the options bar, there's a little pop up menu called fill. And if you ever pull the cropping rectangle out away from the edge of your picture, that's how you could add space and here I could add space on, let's say three of the sides that are here. Then up there, if Phil is set to generative expand, then it's gonna end up using what I just showed you to figure out what should appear in those areas that are being added. So if you had a vertical photo and you needed a square one or you needed horizontal and it didn't work to crop it and actually throw away information, we could use generative expand. So I'm just gonna hit the check box here. Let's zoom out on this picture now that it's done and take a look at what we have. Now before I get into this, there's this little bar at the bottom and that's known as the contextual task bar. And I personally hate it. And the only reason I hate it is it's pretty much the only thing that covers up my picture. If you look at all the other panels in Photoshop, none of them cover my picture. And so this thing, I just always want to move it out of the way. Now you can put this somewhere on your screen wherever you want just by grabbing the bar on the side and maybe I could put it up here and then over on the side, there's three little dots in there. You can say you want to pin it to this position and therefore it always show up there or you can reset it, then it's gonna show up in a default location or you can completely hide it if you don't want to use it. So if you choose hide bar and you want to get it back later on, just go to the window menu, it's found near the bottom. It's called contextual task bar. And I just really wish they would have implemented that in a different way where it wouldn't need to uh overlap my picture ever. Now take a look at these results. It's pretty darn interesting. Let me turn this off so you can see the original cropping that we had and there was very little information to give it a hint as to what should be next on the sides. But it has an idea of what straight lines are like. It has an idea of patterns and how they should continue and other things, it's been trained on what buildings look like in general. But it usually also produces some generally generic information. If you look at the area here, it added a nice little arch, but the detail above it looks like it shouldn't. There is some unreadable text down here and some of the other details above aren't all that great. So I either need to select smaller areas to tell it what to do up here and I could type something in to actually give it a description uh or use other tools like the remove tool to clean this up. But remember you get three versions every time. So let's try the second version that looks pretty amazing. But once again, the central portion I really like, but look at the windows that it added just outside of that central portion. I don't know what that is. Uh So that's just where you could select this area and do generative fill again or you could select one area like here and you can actually type in a description of window or something. Uh And then here is our third one. And as far as those goes, I don't like the bottom of this one. And let's see. Mm It's really hard for me to decide between this one and this one. I'm thinking this one. But let's say that you wanted parts of both, maybe what it is is you like the bottom portion, the way it dealt with this base of the sculpture in this version, but you like the sides in this version. Well, what can you do because they're stuck here in one layer. Uh Well, if I duplicate this layer, I could just type command J that means jump to a new layer. Now, I have two copies of them, both of them are identical. So I could switch one of them to the other uh version or let me choose undo to get back to where I was when I was like this with a single layer. The other thing I believe you can do is right click on this. And I think there's a choice in here right here called convert to layers. And so therefore I can choose that. And now what this is telling me is the image I'm working on is in 16 bit mode and it looks to me like generative fill is generating an eight bit image and they're warning me about that. Well, regardless if I put it on separate layers or not, that would be the case. So I'm gonna choose, don't show again and I'm going to choose, yes, I want to proceed. And unless I plan on doing radical changes to the brightness or color in the image, it's usually not gonna matter. So when I tell it to generate uh layers, then I'm gonna end up with one layer for each one of those results and it's gonna put them in a group and do some other stuff. But uh therefore we come in here and say, well, this one I liked for the side but it didn't like the bottom. So therefore, where's the one I like for the bottom? Let's see. It's not that one. No, it was that one. This one. So I want to use this one just on the bottom. I could decide to drag that up on top, turn the one on that's underneath. And I'll add a layer mask and on that layer mask, I want to fill the entire layer mask with black. So there are many ways of doing that. I'll choose invert and then let's get a big brush and I'm gonna paint with white, white, makes part of the layer show up so I could paint right up here until it makes contact with that and then decide exactly how much this bottom portion I wanna use. Maybe I like that railing going over on each side and then I could use something like the uh tool up here that gives us selections, the object selection tool to see if I might be able to get it to give me a selection of this area up here that I don't wanna include. And I can end up filling that with black. I'll use a keyboard shortcut that fills in my background color command delete. Uh Then I could do the same thing on the opposite side, give it a sense for what I want. And then command delete means filled with background color and that type of thing to mix more than one version of that together. But let's try that on other images to expand using the crop tool. This is a very low resolution picture that has very little information in general. But let's see what it comes up with. If I tell it that I would like to add all this space to the image, I'm just gonna press return or enter. Look at that. It gave me shorts, gave my wife a bikini and it built her arms. Remember how much there was in the photo, I'll turn this off there. You couldn't even see our arms, but it built up entire arms. Now when it does this, it's usually terrible at fingers. So be judgmental of the fingers and consider retouching or trying again and again because like this finger looks like kind of the width of a finger and a half and my thumb looks rather small. Uh her hand, I can't tell that this is like an extra finger out there or what not very well defined. But remember it gives you three versions. So you might as well click between those three and just so you know, you can control the size of these little thumbnails with this little icon up here. So you can either get small ones or big ones. Let's go with the small ones for this uh image and I'm gonna click on the next one. Now. She's instead of wearing a bikini, she's got something else, but there's something weird going on behind us and then I'll go for the next one and that's even weirder. Look at my legs and it looks like I'm just kind of disappearing in the sand and look at Karen's hand that is going into a non-existent pocket and that kind of stuff. But it might be that I like one of these, uh like this one and I just need to work a little bit more. So what I might do now is just say, well, I like the hand except for this part, although this does not do great with hands. So I do not expect it to do an amazing job if I hit generate. I consider that a bit better, but my thumb still looks a little skinny. Let's try the next one. Terrible in the next one. It puts something in my hands. So I'm thinking that first one, I could always make another selection just around my thumb and generate it. Uh Here, I don't know what's going on with her hand and that area just outside her hand. I don't mind it up to about there. So I might make a selection like that hit generate. Now, this is an issue you're gonna run into and it's anytime you work on an area that is skin adobe in their guidelines says you're not supposed to use this on uh blatant nude photographs, that type of thing. So if you ever have an area of skin and there's any detail within it and it's not a face, then it will often uh reject it and it just says we can't do it. It's outside our guidelines so I could come up here and try to type in uh a prompt. Maybe I just say fingers and then at least it knows what I want. Nope, not going to do it. And that's because it thinks it could be a nude or something else inappropriate, doesn't know what that detail is. And so it's not gonna do it. So what I might need to do instead is just generate more versions of it and see if she ever has her hand in that same location or uh find a hand from another picture and put it in here. But if you look at this, it's really kind of amazing that up here uh behind Karen's head in the original picture, it was blown out with no detail. So it did the same treatment down here. And if you look at her neck and the side of her face, you see how it's kind of blown out with the highlights coming over and it did the exact same thing on the arm and invented and on the side of her body there. And so I'm really impressed with it being able to figure out the direction of light sources and things like that, what it will be terrible at our hands feet and anything that has text, uh it's, it's never knows how to make text, but let's close that and move on to another. So this is one that I assume it will have an issue with. In fact, let's just go and use our little crop and let's say add this, the thing is it doesn't know is she wearing a bikini or should she be wearing something else? And it thinks I might be requesting something inappropriate. The number of times I get this result is so high that it makes it less usable. And the other times I tried it with this image literally an hour ago, I did this image almost exactly the same and it gave me the the same uh warning and this time it works. So I'm amazed. Uh anyway, here are our three different versions and wow, I think that looks pretty darn good. Go back to the uh crop tool and continue just, hey, I want to see your feet or whatever is gonna be above. You don't keep doing it. But we should be aware of is the largest, it's able to generate something is pixels square. And if the image you're attempting to do this with is larger than that, then it generates something that's 1024 pixels. And then scales it up and therefore you gotta zoom up and look close because this will likely look very soft and that's because of that being scaled up. But let's see if we got a better one in the second one or in the third, I'm thinking the second one is the best out of those. So let's zoom up and let's look at where her face is. You can see the kind of detail. Well, at least up here it was out of focus. Uh But it only went up so high only to about here. And I can tell right where it kicked in up here, I see a different kind of texture and it just doesn't quite uh look appropriate. And so that's one thing you'll notice. And if I go from her outfit, let's see where this started. OK. It started right there and I can see a little bit of a line there. Uh But oftentimes the texture will be softer. Now, sometimes you'll need to add noise and other times you'll just need to expand things in sections. If I did a little bit, 1000 24 pixels and then a little bit more and a little bit more, then it was going to generate a higher quality result. If it's a large image, here's another uh this would be the original image and I decided to expand it on both sides. I thought the right sign it did. OK. With, if I switch between the three versions. Some of them are acceptable, but you do need to be critical if you look close at this area up in here. I don't know that it's good on any of them if you truly evaluated it. Like, what is that thing in there and over on the left side, I don't think any of them are acceptable. Look at what it does here. We had the, this kind of gate and it just kind of thought it should look a little bit like it and not fully. And when I go through those again, for Instagram, when it's really small, it might work. But for high quality work, oftentimes it won't. Now, let's look at what we can do when we actually type something in to tell it specifically what we'd like. Uh Here this is actually at a wedding and I'm not into the hat. So what I'm gonna do is go to the object selection tool and I'm gonna see if it will recognize that hat and it sure looks like it does. I'll click there. And if the selection is not perfect because it included too much of his skin, I can hold down the option key, that's all and windows and circle around something to say, oh, I didn't mean that part. So now I got the hat, but I don't want a selection of the hat that is perfectly touching it. I do want a little bit of space beyond because there can be a shadow in here. And if there is, it's gonna think it should be kept if it's outside of that selection. So go to the select menu, there's a choice called modify. And that's where I find a choice called expand. And the number you type in will depend on how high resolution of a picture you're working with. If it's a low res picture, you need a relatively low number, I'll type in 10 to see what it gives us and that's gonna make our selection larger, but keep the same shape. So when I click, OK, now, if you were to look close at that hat and I choose undo before, after there's a little gap now between the selection and the hat, but then the shape of the selection I make is gonna have a huge influence on uh what it ends up with. If I tell it to make a hat right now, it's got to fit within that shape. So he might get like a little beanie or something in there or another ball cap, but I don't think he's going to get uh a larger hat. So what I might do is hold down the shift key to add my selection. And I might come down here to say, I think that hatch should go down to this far. And when it comes up this way, I think I'm gonna give it a chance to go up higher there and I'll give it a chance to come down and maybe get a chance to have a little bit of a rim there and maybe even add a little more there. So it just has the, the chance to fill that. Uh Then I'm gonna go to the edit menu and choose generative fill in this time. I'm gonna say hat. Now, if I just ta hat, I'll have no control over its style. So I'm gonna say elegant since it's a wedding. Um men's hat and see what it comes up with. That ain't bad. I mean, I like that. It gave him his hair looks about the same going into it and all that. But remember you get three versions. So over on the right side, here's my second version and there's my third version and I didn't even tell it what color to use or really what style to use. But I'm thinking that that one is looking pretty darn nice and when it does, it ends up trying to get the lighting just right. So it notices this really bright area in the background and it causes that to look like it's reflecting up in lighting the edge of the hat. So then I can come in and let's say now I want to change its tie. The other people at the wedding were wearing a dark uh like maroon kind of a tie. So I'm gonna again go to my object selection tool. I'm gonna come over here. And as long as I have the choice of sample, all layers turned on, then when I hover over this area, I should be able to see if it recognized that object, it looks like it did. And so I'll click there. Uh Then once again, I want to make my selection larger. So select, modify, expand. And I didn't mind before when I used 10. So let's see about that. All right. at least now it is not perfectly touching the time. Instead, it is a little bit of a gap. Now, I could come down here and I see the shadow for this, I can add shift and use the lasso to just kind of include that because if that shadow is bluish, then it's probably gonna put in a blue tie because that's what would need to be there to cast that color of a shadow. But I think now I have all the areas that would be influenced by that. Now, so far, I've gone to the edit menu and that's where I chose generative fill. But if you have this bar, that contextual task bar that was down at the bottom of my screen originally, right? There is a button for generative fill and I'm gonna click that and right in the bar, I can type in what I would like. And what I often like to do is instead of typing in there, uh is I can use my computer's operating system and I can have it, uh, listen to my voice and type in the text that I use uh, on my computer. I've set up a keyboard shortcut for that. I press one of the keys on my keyboard twice and it, it doesn't. So I'm gonna go dark maroon tie and I can use the arrow keys here. Uh I might just add necktie just to be sure it was a little bit still listening to me talking. Uh Let me try that over again. Um When you click there, you might wanna select on hit delete because it left the original text that was in there, which was just telling me to type something in. So now let me try it dark maroon necktie. I don't know about you but to me that's not dark maroon. Um that's more purple. That would be the closest if you don't like what it produced like here. I see some texture. It almost looks like a leather tie. Uh You can change your prompt up here at any time and then you just hit generate again. And so I might say something like solid, meaning not patterned. Um And I'm gonna take right maroon and I'm just trying it again. It's a little bit better in the color, but it's got some texture to it and a little more purple and all that. Uh It's a matter of if you can decide what you truly want. Now, one thing you could do, but I can't do due to copyright reasons and videos is you can search the internet. And if you search the internet for the tie you want and you find it on a simple background, like a white background, there's actually an icon right here where you could feed it a picture to say, make one like this. But the other thing we can do here is if I find that one of these is the best, let's say it's this one here. If I go to the upper right corner, you see the three little dots, one of the choices there is generate similar. And that means give me a variation on this exact one. And therefore, I'll just get a few different versions that should look similar. There we go. And I actually kind of like that one. If you saw the ties, the rest of the wedding party were wearing it, it wasn't too dissimilar to that. Maybe just a little bit darker. Uh Remember once you have these, if you're not gonna use them, you can hit this little uh trash icon here and that is going to delete them. Therefore, I can just keep the one I liked. Let's see if we could change my shirt. Uh I'll use the object selection tool, click on the shirt and then I wanna expand it zoom up to make sure that it goes outside. And it looks to me like that was more than I needed. I think I typed in 12. This must be a lower resolution picture. So I'll choose undo and just head back in there maybe use about half that maybe use five. Yeah, that seems to now be outside the shape of the shirt and therefore should include uh any little hints of that color that were there. But I'm gonna look all the way around the edge to ensure that that's the case. And I noticed on my neck it's selected part of my neck that I shouldn't have. So I'll hold down the option key. That's all to windows and use the lasso tool to remove that part because I don't want it messing with and replacing a portion of my neck. All right, then let's this time, go up to here. Hit generative fill and this time what I think I'll do now, I told it to a yellow and blue shirt. It did yellow black and I told it to be a button down shirt and it gave me a different time. Yeah. There we do have blue and yellow, not button down though. And once again, so I need to somehow decide, is there a different way I could mention button down shirt? Uh I might say collared button down shirt and get rid of casual and it seems like it didn't really do anything like what I wanted. Now, I did yellow and blue stripes because I thought that would be a hard thing to do and it's done fine when I wanted to look something like a t-shirt, but not in this particular instance. There we go. I put elegant semiformal and then it started to do a little bit what I wanted. Although, uh, I don't see buttons on the shirt, even though it has a collar that's open to make it look like that. Let's try a second one and let's try a third one. I like that. I'm looking like a referee. Uh I'm gonna go with that, but I'm just giving you the idea that it can make clothing and that clothing will conform to the shape of the subject and will tuck into belts and all sorts of uh similar things. In this case, what I wanna do is replace the background. So to do so I should get the windows selected as well. So I'll go back to the object selection tool. I can just hold down the shift key and I can trace around what it is. I'd like it to select and so I'll show it to get that area of the window. Although I need to hold down the option key to take away because we already have the car selected. So that took that away. Let's take away the back window, at least this lower portion and then the upper one and then we have a window in here. We can trace around and see if it can find and it did an ok. Job. And let's say that's what we want. And I would like to now replace the background. Well, we're missing a tail light here. So I'll hold to add it and we'll make sure we get around that tail light. So it's not discarded in this time. What I'd like to do is actually make my selection smaller. So it goes into the edge of that object a bit. So I'll come over here and choose contract and I might type in a number like maybe four. So now it's overlapping the inside of that car so that if I choose inverse to get the opposite of it, now we have the background selected, we have the area inside the windows selected and our selection overlaps the car a bit so that you're not gonna get any weird transition zones. I think I could have gotten away with contracting that selection a little bit less, but it might still be fine. So at this point, I can hit generative film and I just need to tell it what kind of background I would like. Then of course, we have three variations so I can click between those. I don't know why it thinks we should be crossing the road, like blocking it, but this one looks like it's driving down the road. And what I would often do here is I ended up making it. So my selection overlapped the car like was not going all the way up to the edge just to ensure that all the edge pixels were included. Well, sometimes I'll take the original picture then and I'll just put it on top and then I'll make a new selection that is precise and add a mask. And therefore if it went around here and replaced any of the parts on the end on the edge, like over here, I think it ended up with hint of blue if I re Yeah, right there that I won't end up having that. So I'll come back to my object selection tool. I'll come back here and get a more accurate selection and then I can add a layer mask and I would refine that layer mask. But by doing so, if you look at the back portion of the car and you compare it to what's underneath, you'll see that there's more car in there and uh it will usually look better. I would just need to refine this further. Then there's a new icon at the bottom of the tools panel and it's this one right here that will generate um an image from a prompt, but it is going to replace the entirety of this image. Even if you had a selection to limit where something happens that still feels the whole document, it'll create a new layer. So it's not gonna replace anything that's in here. But you can start with a brand new document that's simply empty. It doesn't need to contain a thing and click that. Let's ask for it to build a, let's see, 19 thirties typewriter with chrome ringed oops, round keys. So I'm being relatively specific about what I want and then I can specify some other stuff down here, but I don't have to, all I need to do is type this in and now it's hit generate and it's gonna fill the entirety of the image and overall the keys look OK? The space R looks a little odd and it's obvious it has no idea what an alphabet is or what order keys would usually be in. But we have three versions of this. And so we see all three, but I noticed all three of them are zoomed up on the keys themselves. So I probably need to change my prompt to get it to reveal the entire typewriter. So I'm gonna add to this, I'm gonna say, with a sheet of paper loaded sitting on a wooden table. There we go. Now, we're getting the entirety of it and though generally really wants to crop in type, but we have three variations. Here's the second one and here's the third one. Now, it only has a generic idea of what a typewriter really is. Uh I actually have a typewriter right here. That is, would, you know, be exactly what I'm asking for and like these little covers here always have a certain look, they do a specific thing, but the artificial intelligence has no idea why there are little circles here. So it's gonna put different stuff here that thinks it looks appropriate and it's just not going to get everything right. So, if I look at these and am I a little bit critical of them, um, this, uh, this area right here is usually something that only sticks out, uh, a little bit. It's to grab your hand and do the carriage return. It would never stick out that far. I don't know what this thing is and these do not look appropriate the keys themselves like in between them looks rather nice, but it has no idea what letters are, what order they should go in. Uh And yeah, in general, let's, uh, click through the other ones like this one that looks more like a microphone or a something else. And why is there a little, um, thing of a jigger hanging down in this? These are supposed to be the actual letters that would come up there and hit the, the uh ribbon that's there and you can't make out a single one of them. Not sure what that is or that. And it just is odd. Uh, go to the next one. Same thing. Why would you have something that looks like this on a typewriter? So you'll find, if there's complex machinery, it's not gonna go do a good job, it'll be able to put in something that looks generically like it, but not something that is specific where you can look at the details and actually understand it. There are some things, it does far better at animals since they don't have to have, you know, you wouldn't notice if there's something the littlest bit wrong with some part of this. Uh you might be able to get away with. In this particular case, if I click on this, when I have my variations up here, uh I can click on any of those variations. And up here, I can read the prompt that I ended up using to generate it. And so I can see that this one, this one and this one, I use the same prompt because you get three versions when you do and I wasn't as happy with it. I wanted the, the eyes to have a little more intensity to it. So then I added, I said with concentrated fierce looking eyes uh and filling the majority of the frame to make sure it was large and the eyes just don't feel uh if you have a lion looking at you, I've had it happen in real life. It uh isn't quite like that. I think a lot of these are pretty good, but one thing I could do is there's an icon right here and that means if I click it, that I can supply it with a reference image down here, it gives me some reference images. I could click on to tell it. I want an image that looks a lot like one of these, but rarely am I gonna find anything in there? That's truly what I was looking for. So instead I can choose image here. And if I do it wants me to open uh an image and I happen to have a picture in here, I think it might be this one that is actually of a lion, I'll choose open and now it's in this little spot here saying that it's going to investigate that and evaluate it and use that to influence uh what the results will be. So I end up closing that and then I'll hit generate and we end up with these three versions. And I can see up here that I am using my reference image and I do like the eyes on these. They have the look that I'm used to with lines, especially this one here. Uh But it is doing it based on that reference picture. It's not actually using any part of the reference picture. It is just looking at it, figure out the style and content of the photo that I'm looking to generate. If I decide that it's not helping, then you could go over here and click on that same icon and there's a trash sign right here. And if you clicked on that, this would disappear and you could generate once again and you'd no longer be using that reference image or you could hit, choose to choose a different one. But then let's create a brand new document. And when you create a brand new document, if you wanted to fill that document with something that you type in a prompt for, you're gonna need it to have something in that document, uh white black, whatever, you just can't have it such a transparent otherwise it's not going to work, but it's as long as I have something like this, then I can go over to that icon on the left side. Remember this is only in the beta version currently but should hopefully get it to the main version soon. And therefore what you see in the screen is not in the actively shipping version of Photoshop uh but should get there soon. Now over here are some images that have been generated from a prompt by typing in text and we can be inspired by them and also learn from them. So let's click on one. And when you do up here, you're actually gonna see the prompt that was used to generate that particular image and you'll start learning things like asymmetrical balance, meaning I don't want whatever it is to be centered and symmetrical on both sides. Uh You can say what type it is, is it elegant or and the more influence you give it, the closer you're gonna get to what you were really looking for. So let's generate this. Now it's not gonna generate this exact image, but it should be somewhat similar to it because it's only looking at the text, it has no idea what that picture actually looked like. Uh kind of interesting because this is symmetrical and specifically within the uh instructions, it said asymmetrical balance. But if I click on the second image that's asymmetrical where it's not lined up perfectly with the same information on both sides, it looks pretty darn nice and then the next one and we can just cycle between them but be sure to be really careful and evaluate it. Because if you look at like the edge of this countertop, I can see that suddenly it doubles up right here and then it goes thicker right there. And then it's, I don't know, it just doesn't look appropriate all the way across. It might be that the majority of this image is usable, but we need to look at those little details also right here on the edge of the window. I'm not sure what the heck's going on there. So what I might do in that case is create a brand new layer to put above it so I can apply some retouching. And I'm gonna use the remove tool. The remove tool is very good at maintaining straight lines and trying to maintain thicknesses. So I'm gonna come over here and paint across the edge of the countertop in the hopes that it may be able to make it more consistent. I should have left a little bit of it behind if I wanted to specify how thick it should be. I think I did leave the tiniest bit at the end, but now if you look, it's consistent across there, although we did lose a little handle. Uh, so I would come in here and clean up any areas that just look odd. If you look at the floor, for instance, what's going on here in the middle, it just seems to not know. And oftentimes that'll happen if there should be a reflection or something, it just doesn't get the general concept. Uh I don't know what this is on top of the table, but I can probably use the remove tool although I left a little hint of it there uh to fix that. I don't know what the heck this vertical thing is. Uh so I can go there and it's OK if you overlap things here, I'll try to get that corner to be cleaned up. It seems to not really understand that. And so you do really need to evaluate this afterwards and you will likely need to touch it up like this back leg seems to go to nothingness up there. So you'll either need to copy another back leg uh or end up doing some retouching or select it and say wooden leg uh something like that. But let's look more at how you can be inspired about and control the results you get. So let's click on the icon again. Remember, you can click on these and you can read what the description was that was being used for each one, you'll find some of them look more like drawings than photographs or just look like kind of art like this. Well, right here is a choice. You can say you want a photographic version or more of an art version of whatever it is, you're prompting it to create. And so you notice when I click between these, if I click on one that looks like a photo, it was set to photo. And if you click on one that looks like it's not a photo, it was probably set to art. And then also down here, we have effects. And if I click between these, you might notice a number appearing there. Like here, there's a number one. And if I go down here, there's not, well, you can click here to see what kind of effect is specified. When I click, there are all sorts of choices you could choose in here. But if you have the set to select, then you're gonna see what choice was in that prompt. What one was used with it? Well, what we can do is here, I'll clear all and I'll choose the choice of all. Now, I can see all the different uh choices they could use for effects. They are categorized into these sections. So I could say that I would like something that, but see for a particular technique, this would usually be a kind of painting. And I could say I want, let's say a watercolor painting. But then I could also come in and talk about themes. Do I want it to feel very cinematic or do I want it more comic book, you know, that type of thing? And this really influences the look of the end result. So maybe I'll come in here and I'm gonna choose cyber matrix and then you can go to other sections here's concepts. So in this case, I want it to look futuristic. You don't have to choose any of these, but the more you do, the more you influence the look of the end result. So I'm gonna generate this and we're gonna see what we get. We're asking for an old lighthouse at night in operation and all that stuff made out of a layer of paper, we get three versions of it. There's 1st, 2nd and 3rd and it did mention layered paper in the description. And if I go back to this little icon here in my options, the reason why I didn't end up with it is I told it to be a watercolor and because of that, that kind of overrided some of the description that was in there, I could always turn off watercolor and then just hit generate again. And here are our three. But I think that the choices I have in here are largely overriding the uh layered paper because I'm not sure how layered paper is supposed to look futuristic in cyber matrix. So I might turn those off and finally hit generate to see really what it was intended to create. Let's look at one last thing. Here's an image that I selected the background on. And this is a selection that I used. Then I went to the edit menu and I chose Jit to Phil and I told it to put this on a country road at sunset with some trees in the background. And it ended up producing this image. Well, when you go over here into the properties panel, if you use generative fill, instead of using that little icon down there, then when you hover over one of the variations that are here, you will likely find in the upper left corner, an icon right now that an icon only appears in the beta version of Photoshop. But hopefully it'll make its way to the main version soon. And if you click it, what it's gonna do is scale up the version of the content that it generated and it'll become twice as big. So if I hit that, sometimes it'll tell you that the area is too small to have it apply to. And that means that the area that it generated was 1024 pixels or less. And when it does do that, you're gonna end up with an extra generation in here. So that if you zoom up on your image and you switch between those two, you can see what it looked like when it originally was created. And then you can see if you click on the next one, what it looked like after scaling it up, it did not regenerate the background. It instead used the equivalent to uh going up to the filter menu and choosing neural filters. And then there's a choice called Super Zoom and it can be helpful if you end up needing to replace a large area of a background. Well, I hope it gives you a little bit of a sense for how to use generative A I and Photoshop. This is a fast progressing technology. So I would assume that you'll see a lot of changes uh and also the quality will improve relatively quickly over time. So a year from now, I'm guessing what we're gonna be getting is gonna be a lot better because I've seen improvement in the short time. It's been in there.