Image Control Through Prompting: Mastering Camera Perspectives | Alchemy with Xerophayze

[Music] hey everyone welcome back to Alchemy with zero phase this is Eric and I wanted to take a little bit of time to talk about uh prompting but more specifically controlling like your view in in your images through prompting um I've looked at a lot of people's responses like on Facebook I get on a lot of these forums uh stable diffusion forums uh or even just AI image generation in particular and I noticed a lot of people having the same issues I had a year and a half ago when I first started doing AI image generation where the like let's say I'm working on a character and they're like half cut out of the scene or they're not I can't get the full body in the SC and um over you know this time that I've been using this uh you know I figured out a lot of techniques and being able to control that as you know that aspect or just controlling the overall scene in general through the prompt um without having to say things like uh full body head to toe or um vast scene or something like that uh you can control the view of the image through a couple different aspects you got to realize these AI love details and when you put details into something it focuses on that the AI is going to focus on whatever it is you're putting details into for example let's just do a simple prompt here we're going to do a um man we'll just do let's do a photo oops photo of a man looking into the camera okay and I'm not giving much detail it's a photo it's a man he's looking into the camera so my expectation of this I mean aside from creating an image where the guy's like holding a camera and he's looking into the camera which I don't think it'll do um should be a very close-up shot okay um of just his face uh and just looking directly into the camera and nothing else let's go ahead and set this I got a preset for portraits here it's going to change a lot of these settings um we're going to shift this here's the other aspect there's two really big aspects uh that uh settings that will control the overall uh composition of the image one of them is going to be what you're detailing what you're describing the other aspect is really going to be the aspect ratio your you're utilizing here now I can select 5x4 and it'll create a man kind of centered probably looking into the camera but when I change it to something a little more portrait 4x5 or a four five aspect ratio the AI has been trained on images that that's tends to be more portrait that's going to be it's going to it's going to enhance or um help the image be more what I want uh which is going to be a man's face looking into the camera so let's render that out real quick yeah that's pretty much what I expected it to to show so you get kind of this closeup image if I rendered that a few times it would probably give varying degrees of closeness to his face now let's say I do this look into a camera um brown eyes with oops eyes with white eye lashes and green irises I just cannot spell today now here's what I'm doing I am giving more detail to just the eyes and the the features around the eyes and what that's going to tell the AI to do is let's zoom in see what I mean now I didn't give white eyelashes but uh um it did give them green ey Ires I I was just kind of throwing stuff out there just to give it uh I could say long eyelashes that might actually even you can tell because I'm detailing guys yeah I brought it in even closer because it wants to uh bring bring out the detail that I'm describing um let's do this uh if another word that will help this intimate okay intimate photo it's not what you think intimate means close up you're close to okay again it's very close getting getting out there I didn't expect it to change a whole lot now let's say we want to change this so that we want it super close think of things think of what you're typing in here I put eyes so if I put intimate photo of a man looking into the camera brown eye with long eyelashes and green irises so it's not plural anymore so now I am should get a single eye oh bigger eyes though let's see intimate photo so let's say we get rid of that man's of a man brown eye Green Oh green Iris when you're putting plural in there the the AI really focuses on that see that's what I'm talking about got to get rid of like I was describing irises so it assumes two eyes so now we got ourselves a super close up you can even see the veins in the eye okay and this works in the other direction too okay we can bring this out by changing the details we can leave brown eyes but when you're drawing the image back we are we don't care necessarily about eyelashes okay or the irises we're just going to say brown eye intimate photo we're going to get rid of intimate then we're going to say we let's draw it back a little bit instead of just his face I want his shoulders in it too so um let's say a photo of a man wearing a dress shirt with a bow bow tie let's say a red bow tie and I tend to put phys like the physicality of the person so before wearing something so man with brown own eyes and smile so you're finding a balance here okay I don't want it to bring too much of the uh the the lower body in so I'm adding a few details about his face he's got brown eyes and he's smiling okay we'll put a comma there wearing a dress shirt with a red bow tie now I'm saying a dress shirt I'm not giving it any more details but I am saying instead of just a b tie I'm saying a red bow tie so it's it's going to try to balance out where the camera is focusing yep that's perfect so as you can see I got what I wanted because I was kind of finding this balance of the details I wanted in there now I could say if I added more details to the shirt like a white dress wearing a dress shirt uh with a pocket and and brown belt now that I've given the shirt and something a little lower oh let's get rid of no let's not get rid of that I's see wearing a dress so I didn't do much with the brown belt sure with a pocket I got the pocket let's get rid of again we're finding balance here let's get rid of the smile try that this is experimentation just showing giving you an idea of like really how to do this photo of a man with brown eyes Okay so we're going to say we're pulling back enough I'm not going to worry too much about this get rid of the comma wearing a dress shirt with a pocket and brown belt let's get rid of this right here I want the red tie in there but I'm going to change where it's at with a pocket and red bow tie and brown belt with denim pants most of the details are still up here but I'm really trying to tell the AI I want you to pull back I want you to start showing more of the image more of this person okay so we lost the face so now we have this in here okay which is kind of kind of what wanted so let's come up here of a man smiling with curly hair I want to make sure we keep his head in there so we give it more one more detail about his head there we go that's more like what I was thinking I wanted the belt partially in there but I wanted his head still in there now this is more portrait photography so I could have said portrait photo okay give all those details and it would it would probably have helped it but what I want my goal in this video is to show you that you can do a lot of this through the details of what you're describing without using some of those in there so now we're going to shift to Bringing more of him out okay we're going to zoom back even more so now we've described brown belt with Den in pants um rips in pants army boots so we're 32 tokens out of the 75 that you really kind of that the AI is able to pay attention to okay there we pulled it back even more it's given it a random background we lost the bow tie okay so if we rendered it a few times like let's let's say we shift this up to two renders let's just see if we can keep the bow tie it's giving him more of a casual look because of what we're describing so we got the bow tie but we don't get the boots I'm going to shift this now the aspect ratio to a 9 by 16 the AI is struggling to be able to fit this in there now I could I could draw this back but what I would end up doing is probably uh describing more of the background adding maybe another detail to the boots just so we can get in there like we could say um black army boots so let's shift this back just to see if that works not quite back like that one was close so let's do this so part of stable diffusion allows me to emphasize we're just going to do a slight emphasis on that should draw the image back a little bit you I wonder if it's having a hard time with bow tie because maybe I'm spelling it wrong is it no it's bow tie it should be what it is b o WTI E I think it's hyphenated let's give that a try yeah maybe it is I don't know it's like that one worked it's not black army boots see we're starting to lose focus so probably I will get rid of smiling cuz here's the other thing you need to put Priority on the general aesthetic and the overall layout of the image so don't worry about like when you're pulling the image back like this don't worry about the small details you can go back and in paint those you like like I don't necessarily need to say he's smiling okay I could just say brown hair but I'm going to leave curly hair wearing a dress shirt we're going to get rid of pocket ripped oh oh rips in pants ripped pants I'm going to say ripped there we go that's better that's a little more what I was thinking I this guy's just got some style issues I mean obviously that's got a tie there we could pull that out with in painting super easy again with aspect ratio this actually helps I could do this and it would give me the guy standing there probably every single time without cutting anything off and it balances it out better too lost the red bow tie so we're just going to say tie I don't like having to focus on that too much so now you could actually focus on like a negative prompt at this point so um if there's things in the image you didn't like like brown boots you could start saying that and that would start to solidify more of what you're looking for so you start getting more black boots he's a little more stylish okay now to really pull it back so we're going to we're going to forget about some of the smaller details here I want photo of a man now we're going to say we're going to use something like this we're going to say wide angle photo of a man with curly hair and now we're going to say with a dress shirt red tie brown belt and denim pants rip pants black arm Bo Okay so so now we're going to start focusing on the surrounding environment standing on a Rocky Hill now we're giving the environment more detail okay so we didn't specify we're just changing the background now we're going to say um large Forest cloudy Sky we're just adding details to the background and the other thing we're going to have to check see it pulled it back a little bit there yeah so now the it's doing a really wide angle but in this narrow aspect ratio but we're getting more of the image in there and he's looking better stylish so now at this point we're pushing the limits of what um we could do here now we could add more details about where he's standing okay standing on a Rocky Hill I mean this aspect ratio is really going to elimin us but let me let me try something here on a Rocky Hill um next to let's say under under a Ancient Temple so we're giving it something large to focus on an ancient Temple is Big right so yeah it's drawing it back even more it's still doing this wide angle thing okay but it's now pulling him back and it switched shirt in the tie it's having a hard time focusing on that but that's okay so as you can see we're pulling it back even more now if you really want this to pull back we're telling it's a wide angle so now we need to shift the aspect ratio over to something like a 5x4 so it's a little more wide than it is narrow you know wide than wider than taller so I don't know however you want to put that so the aspect ratio is wider than it is tall and if we render this we're getting something that looks a little more natural okay pulled it back we're getting more scenery here okay we could change the aspect ratio again to a 16 by9 what's cool about this is because we're describing enough of his details it's keeping him fully in the scene okay and let's uh shut that off we're going to make we're going to change the width on this to 1280 by 576 this will give us a really wide shot let's see what it does with him puts see them right in the center on that one but uh this one here so it's almost like we're getting this really cool wide angle cell phone shot um obviously we're losing detail in there I'm not worried too much about that but uh we're still getting the details we want we got a cloudy Sky now we could say um I mean at this point you're just adding details in and you can make that as wide as you want okay and it's going to keep him in there so uh now if we really wanted to pull this back this is where we're going to start introducing different uh camera angles so what we want to do we're going to do a high angle photo I want to see what that does high angle I can't that was like from below looking up or yeah no it's it's from the top kind of looking down yeah so you can kind of see the angle shifted up a little bit now let's do um drone shot get rid of angle it knows what a drone oops drop shot that would be interesting drone shot think I spelled that right this time it's pulling back even more and yeah it's getting a much higher perspective here's what it's doing though because we are really focusing on the details around this man it's keep him in there even though the angle and everything you're getting a wider view of what's around him as if you were a high above everything because we told a drone shot so if we were to change this we're going to say photo of a man we're going to get rid of curly hair and wearing a dress shirt we're going to not worry about the belt we're gonna not worry about ripped because that's just telling it focus on the pants we're just gonna say black boots this may still be too much but standing on a Rocky Hill under an ancient Temple large Forest cloudy sky so we're reducing the number of details around the man let's see what it does little higher angle it's a little smaller let's see let's do drone shot photo of a man yeah we're going to need to just basically back off some of these other ones cuz a lot of these we can just in paint so of a man bring your dress shirt denim pants boots standing on a Rocky Hill under an ancient Temple large Forest cloudy Sky see if we can get that to pull back it may just be we have to say image of a man oh there we go that's that's better we're we're tell you know because we haven't specified anything about the man's face it's some of these images he's going to be turned away let's do a bird's eye shot not sure if I'm doing that bird's eye if it's one word or two words let's see what it does now that pulls back even more gives you much wider angle here now there's lots of different camera angles these are fun um I am referencing a lot of the data I have in my my prompt online prompt generator gives you access to a lot of that um you can do point of view shot selfy shot selfie is kind of a fun one let's do this selfie it's wide though normally those would be yeah let's shift this back over to a 9 by 16 because most selfies would be vertical right selfie shot let's do this man taking a selfie sometimes it's how you word it too like the action of like what is he doing something I think I'm doing here is uh yeah so let's get rid of this right here I'm not liking that there I mean that's kind of better we're coming in closer it's a you know it's trying to do the selfish shot hey I have a hard time with this sometimes um with uh figuring out it's supposed to be the camera um I'm G to try something here give me a second no no no no selfie photo selfie photography okay let's drop that in there sometimes it is about like what it is you're trying to like how it's described in the prompt yeah it's still not doing I'm giving up maybe too much I've had a lot of better A lot of better success with selfie photos maybe it's the type of photo I'm telling it to take not sure anyway so to kind of move on we can do overhead shot we could do macro shot mirror you know uh what we got here um high angle low angle let's do low angle low angle photo and as you can see it kind of goes down that is interesting uh photo yeah I'm not specifying it looks like it lost that so man wearing but we got the low low angle photo you know some models are trained on like even really low stuff like like we could say um what is it called worms I view photo not sure what that's going to do yeah it didn't do anything not what I was hoping to do we can do perspective extreme low angle perspective itive photo sometimes you can mix the the uh words and the phrases together no it's messing it up let's get rid of this whole thing extreme low angle photo of man standing next to Temple Ancient Temple [Laughter] that's an extreme low angle photo yep the low angles are all about like putting the main subject really low so you get more of what's above them so kind of fun to mess with so some of the ones I've worked on before are macro shots too and getting it as small as I possibly can and I actually created one of those and turned it into a style called drop inside macro water drop so we're going to do man in hiking gear standing on a rock see what it does see if it does what I hope it does yeah so I set this one up specifically to be this extreme macro shot and to put whatever it is I'm describing inside of what I I would hope to be like a uh a water droplet sometimes if I'm putting too many details in here it uh messes up but let me show you the prompt for this it's kind of fun so if you click your editor you go into to styles and you create Styles so with this one here I did an intimate macro microscopic shot tiny miniature and then you put in the curly bracket prompt curly bracket that is where the description that you're putting in into the the positive prompt will go adding to the overall prompt I've done this with a lot of different things um so I know this video F is focusing on adjusting the view um without having to use words that the am I just can't associate with things like if I did um a full body shot of a I don't know uh woman with red hair I don't think it's going to quite give me what I want and honestly with this model it might give me something I don't necessarily want I have to cut that out so I had to make a few modifications and add in at least some details to keep the clothing on the person and a negative prompt so with this here I said full body shot we're not getting the feet though even though we're using a 9 by6 aspect ratio and even this one was even closer um we could say some people have tried full body shot head to toe see if that helps I just don't think it's very not necessarily very reliable so we're we're not getting it there this one kind of I don't know what the heck's going on right there but instead of saying that we just say photo of a woman with red hair white shirt tan pants and black St let and there you go and it's going to be a lot more reliable that way when you give it at least some sort of a detail that tells the a I need you to focus on everything from top to bottom but here are some details to help get you to focus doesn't necessarily know what to do with head to toe or full body shot most of the time when you say body it's going to focus like on the Torso I I think you're not giving any details about the feet or where where they're standing maybe they're standing on gravel you know that'll change the overall environment too so as you can see it gives you something that's going to be a lot more reliable when it comes to camera manipulation and Camera uh just positioning in general um you could do over the shoulder photo this is a fun one gets the person to turn a little bit whoa that's not interesting cut her off completely um with red hair and white hair say looking looking over her shoulder looking back because when you do it like that it's it's kind of it's like okay it's GNA she's goingon to be looking back at you over her shoulder there we go we did we have the gravel there but we lost the um the black stilettos so let's just enhance that a little bit tell the AI hey it's important there we go very cool so that kind of gives you an idea of how to position characters in there as well you know changing their bodies you could even say like instead of over the shoulder we could do um 3/4 turn it should be more like that because that's the 3/4 turn yeah anyway that's it with that being able to control camera angles um the zoom in zoom out based on the number of details I hope this was uh helpful and gives you a little bit more idea of of how to control that you just using the details in the prompt we'll probably come out with a few more videos on prompting and structure to really help people um understand you know how to use prompting to get the result you want so honestly I would love to see some comments in the in the in the uh comment section about like what you would like to see possibly in the next video on like what are you struggling with impr prompting what's what's the most difficult thing you're you're having to understand or trying to do and just having a difficult time doing I'd love to see that so like uh like the video subscribe to my channel uh love the uh the support that we've been getting uh we've got a pretty good uh uh Discord Channel too A lot of people on there using comfy UI which I still hav't do dove into yet and uh just overall having a lot of fun so we'll talk to youall later

Channel: AIchemy with Xerophayze

Views: 1,333

Keywords: Stable Diffusion, Automatic 1111, image generation, AI art, camera control, image prompts, close-up images, distant images, detail prompting, tutorial, Alchemy with Xerophayze, art creation, digital art, AI technology, Forge Edition, instructional video, art tips, creative techniques, visual art, AI software

Id: s2GQOwGp8H0

Length: 34min 1sec (2041 seconds)

Published: Sun Apr 28 2024