190 | Behind the Curtain: AI Image Generation That Actually Works for Business with Ross Symons

23:29Create an image of a beautiful person. Someone might look at it and go, I don't see that as a beautiful person at all. But someone else might look at that and go, not quite my tempo. But it's all about describing physically what, like in your mind, what, maybe someone with massive scars on their face is someone who's unattractive. But someone might find that quite kinky. So it comes down to taste, it comes down to context. So yeah, I don't want to hop on too much about this now i've done the same thing this is just to show that you can add text into images it was something that only up until recently was almost impossible to do and if you ever try to put text in any of these tools it would very seldom come out unless it was a single word like hello or hi or welcome or whatever the case was whether you've used this you know to try that or not take like here for me it was bad it was the text was just all over the place and that's because i think these tools are image generation tools and not text and design tools which you'll see an example going forward now.

24:26So yeah, this is just a basic example of, and I built on the prompt, which is a cozy living room. Now cozy living room, we've got, you know, it is cozy living room. Are they, this one is animated, looks like something from a, you know, story of some kind. Again, because contextually, we haven't told it to be specific about what it is that we're looking for. Then I've gone in to add a dog on the couch. This isn't even a, this looks like a painting or an illustration of some kind same here that looks like a cat so it's not on the couch it's on the floor not on the couch either you know so where i got to here is like i started with my journey because my journey is the worst at putting text into images so i've asked it to say so the prompt was a cozy living room in the late afternoon warm natural light soft shadows a golden retriever sitting on the couch which it got right books and plants in the background realistic textures framed quote on the wall that says welcome home friend in handwritten text now Now, if you can tell me what that says.

25:25The welcome says welcome. Other than that, it's not exactly what it needs to be. It's not exactly. So, you know, and it's tried on all four images. That's one thing with these tools is they usually generate four images because one of them maybe is the right direction that you want to go in. But chances are, it's probably not. And this one kind of got it. I think this is not a bad job. But now the dog is not on the couch. And it's not a golden retriever. Not a golden retriever. It's got a weird head. There's too many plants in there. Anyway, you see where I'm going with this. So that's just what that was able to do.

25:52So if we move forward now, let's go to image effects. Very safe options. I mean, very realistic images, because I assume, you know, if somebody types in that, that they want a photo of what it is they are looking for. So there's a living room in the late afternoon. There's a dog on the couch, golden retriever on the couch. So again, very similar. So it's interesting how in both cases, the more granular we became, the more similar the images became, which I've only noticed that like as we're speaking now, which is just interesting. that's specifically on Gemini like on image effects yes the images almost the four images became almost identical almost identical with from the same angle the same dog and the welcome home sign at the back well welcome home friend which it got right I think it did a pretty good job there I mean I think on all of them all of them are spelled correctly all of them looks like handwritten picture yeah yeah let's start thinking about from a branding perspective if you want to sell a specific couch or a specific dog for that matter or a specific you know you want to sell your art on a wall.

26:48This is how you start thinking about that. Like maybe you don't want to have welcome home friend. Maybe you want to place your image or your art piece or whatever it is, or a pot plant that's trying to sell or a cushion of some kind. I mean, it really does it to for a single use case. This is like, it's impossible to explain, but it is very useful across the board. And then, so that is from a, you know, from an interior perspective, the last thing I want to go through is a let's have a look here I'm conscious of time and anthropomorphic, try and say that 10 times in a row, anthropomorphic fox which is just a human-esque anthropomorphic just means human style fox.

27:24Okay, we've got an illustration which he's got horns which is kind of weird. He's got a top hat or some sort of bipolar hat on and a coat cool looking fox. One of them's you know, again, we've asked it for that and to get more specific we've asked it for a second time. This time he's wearing a cloak okay so now it's the fox wearing a cloak standing in a misty forest again so this is starting to get a little bit storybook-ish we're all we're still in mid-journey here now we've got him in which it didn't listen to the prompt at all looking toward the viewer with a calm expression none of these images are looking toward the camera but toward the viewer which again this is where i think it gets frustrating for people who are using ai for the first time particularly mid-journey because it does require iteration.

28:06It does require an understanding that you're going to be sitting there for a while, especially at the beginning. But the structure that I'm using where I'm building on the prompt, this makes it so much easier for you to go back at a specific point and go, okay, change the fox to a squirrel, change the misty forest to a beach, where at least you're sticking to some sort of format. Also, if you want things to be more prominent, for example, if you want a wide angle shot, put that at the beginning of the prompt as opposed to the end of the prompt. These things that you want, this just works particularly with mid journey really well.

28:33take and i use this with angles and composition so if you have a wide angle shot of a person on a beach or a fox in a forest put the wide angle at the beginning as opposed to the end just a nice little hack treat there so now we've added more details this is getting a bit more real so we've asked for the same fox wearing a green coat standing in the forest holding a golden crystal or a glowing crystal in one hand cinematic lighting 85 millimeter lens so cinematic lighting automatically implies that it's going to be realistic so you don't have to put things in like realism and you know explaining the details that make it look real so this fox i mean not quite fantastic mister but i mean he's there he looks real ish he's he's quite charming he's still not well although we didn't specify that we want him looking at the camera getting to the last section of this we've asked it to look directly at the camera because i've put it at the beginning so i made sure that let's look let's read the entire prompt because that's actually very different than the outside so i said looking directly at the camera an anthropomorphic fox in detailed fur texture wearing a green coat or a green coat with a gold embroidery standing in a misty forest holding the gold glowing crystal in one hand and the cinematic lighting with the 80 millimeter lens and it didn't get it right with all of them but he's a lot closer to looking toward the camera cinematically you're breaking the fourth i'm getting deep here but you're breaking the fourth what they refer to as the fourth wall when the character in the movie looks directly at the camera that very seldom happens so if you bring cinematic lighting in it's automatically going to make the subject look away from the camera.

29:58Yeah. Well, one more thing that I want to say to people. So here we're going for the look of the fox, right? We said that he has a calm look. If you're trying to capture detail of faces and expressions, the best way to do it in mid-journey, and I think it's the only tool out of the four that has it out of the box, is actually to do a close-up of the fox's face and then outpaint out of that. So you can go and zoom out on a picture on mid-journey and ask it to then add whatever it is that you're going to describe. And that's an awesome way to start with a highly detailed fox face because it's the only thing it's painting, like the whole picture, the whole prompt is describing his face and his look and so on.

30:35And then you zoom out and you can add the forest and the mist and all the other stuff. So that's another little trick when you're trying to create expressions or very detailed face of a person or an animal in this particular case, and then zoom out from there. Yeah. Outpainting, I'm just doing a quick example here is you basically, well, this essentially is in painting, but you're painting out the section that you want to change. So whatever you put in the prompt box now will be filled in the background there. Little, very fancy little trick that people have used here. So yeah, I don't want to really want to get into the subject stuff, but just a quick.

31:07So I have an image here, which is, I like the style. Let's say I like the style of this image, which is pink clouds in the background. There's an ocean, there's a, you know, Narnia style door leading from away from a chair. I like this image. I want to take the same fox that I had. I want to use the exact same prompt, but I want to add that style into it. So simply you just add the prompt in, you drop the URL of the image that you're using, which is the image that I had there. And that allows you to, you've got options to put it as a style reference. You can use it as an image prompt. So this is where you start getting real control over what your image is going to get and also consistency.

31:44You're not going to be doing a fox in a green jacket for your brand. Maybe you are. But the idea I'm trying to get across here is this is how you cross pollinate between or you grab images from anywhere on the Internet, upload them and use them as style references to guide your final image. So you can already see here that just using this, just simply dropping that in, it's changed the style of the image entirely. MidJourney has something called sref codes, which are super handy as well, which help you guide the style. So here we've got like it just looks very cool and it's definitely adhered to that.

32:13And you can crank up the, you know, how strong the style is and how much you want. Again, for those of you not watching, just to give you a little bit of understanding what we're doing. In Mid Journey specifically, you can upload an image or use a previous image that you created with AI and say, I want you to capture either the character, like the subject, or I want to capture the style of the image and then apply that to the next image that you're generating. And that's basically what Ross did. He took a beautiful kind of like sunset pink clouds on the ocean and said, I want to use this style with a fox that you have from the other image.

32:49So you're basically mixing the subject from an image with the style of a different image to get a new image that really adheres to both concepts, which is a very powerful tool to get exactly what you want. Because instead of trying to explain with a million words, the style, which is something very amorphic and is sometimes hard to explain, you can just bring an image and say, okay, this is the style that I'm looking for, but I want this character or this product in the the style, and then it's going to do the thing for you. Exactly. And off the back of that, you mentioned like a style or character consistency.

33:18So I picked this character as the fox in his little green cloak. And with the image that I had, I wanted to create a... I basically wanted the fox in a different scene, but I wanted the fox to be a consistent character. So what I'm scrolling through now is the same fox. He's in different positions, but he's got... why I've used the same style across. And this is how you start bringing consistency in elements, objects, sub, but particularly with characters. When you are bringing a subject in, I didn't do it exactly the right way here, but if it's got a clean background, so you just have the character, get rid of the background, Photoshop it, remove the background, make it white, and then it focuses entirely on just the character themselves.

33:57And the more detail you have in the image itself, the more it's going to place that detail into the final image of the character itself. that is just a quick rundown of what's going on there with my journey then the example we have here so an anthropomorphic fox it's done exactly the same thing but here we have no control now you know it's this is like a i don't know these images are kind of cool we've just built on that again there's a something i wanted to point out here the difference between this image and the The other one is purely the aspect ratio. It's the orientation. And that has a massive influence on, this is where understanding a little bit about film and a little bit about how to take photographs.

34:39If you say you want a portrait of a fox or a portrait of a person, place it into a portrait. Don't make it a wide shot. It's not, it understands better. That's a little, just a neat little hack that it's often helps you get the results you want. So we're just building on our fox here. He's getting a little bit more detailed. He's getting a bit more real, definitely adhering to the prompt, but very similar. Like you can see, it's just picked a character. And it's just made him across there very safe. Like this is the character we're going with. So, you know, do with it as you please. If we go and have a look at what ChatGPT has done, we've done, oh, we actually missed ChatGPT with the welcome home.

35:12You can see the welcome home here is perfect. That's one thing ChatGPT. Yeah, so I think from a text perspective, ChatGPT right now is the best tool of adding text to the images. Number two was number one for a very long time, which is Ideogram. And then number three is Gemini. So if you want to get text right in the image itself, go to ChatGPT. you're probably going to get it 19 out of 20 times. It's actually going to get it right. Even if it's a lot of text. Yeah, exactly. So here we've got our Fox and it's made it illustrative. You've added more details and I think it's come out with an image.

35:40That's pretty cool. I mean, again, kind of similar. I mean, a little bit more moody than the one that we've got in image effects with the yellow sepia tone, which is the, yeah, like the petate, which will get changed. Anyway, I wanted to get into, so yeah, I mean, if there are any questions or anything, you're very welcome to speak about that. but okay so yeah the next one i think is awesome so i think now we're going to dive into product photography right look i just this is not a real brand i created i just said i need a brand to you know it's a fictitious brand i came up with the word fleece i why that popped into my head i have no idea but it's a fleece and it's a fragrance brand cool so what i did is i took the image and i cut it out and i said in chat gpt i said cool okay place this is just a workflow from here there are a million directions to go.

36:27Is there a best way to do it? I'm not going to say there is, but this is one way of doing it. Getting your product into a photo is like, it's very difficult to get it perfect. If you want, if you're looking for perfection, then hire a designer. But if you want content that is usable, that on social media and just for quick content, this is one process of going about it. So you cut the image out, make sure there's no background. Okay. I've asked, place this, because now I can speak in normal English language or any language. Place this fragrance bottle on a matte stone pedestal, soft lighting editorial style shot cool so the difference here i mean you can clearly say it's just squash the bottle for some reason it's just made the bottle too small i mean it still has the branding which is great the all the text is very legible and it's a nice image but the bottle is not i mean if you're going to be selling this on you know anywhere amazon or whatever if the product doesn't look the same then people are going to get it and be like what's going on here like this is not the asphalt you get where i'm going yeah two things about this that i've learned that works really well in ChatGPT specifically when it comes to trying to place a product, an actual product in an actual image, two things that work for me very well.

37:28One is I ask it to pull the text off the product and write it just in text. Tell me row by row, what does it say and what kind of the style of the font and how big is the font? And then you kind of get a description in English of what is written on the bottle. In this case, it's just two things, but usually a bottle has, you know, the size and how much liquid it has in it, the percentage of alcohol. If it's a sunscreened and all the different, there's a lot of stuff that's going on usually on a product. And if it writes on its own all the stuff down, usually then you tell it, okay, now use everything you written, write it back in the same style on the product.

38:02It usually gets it right again about eight out of ten times. It's actually going to get the text exactly right. The other thing that I've learned, going back to the same thing, is ask it to describe the product in detail, including color palettes, specific color scheme, aspect ratios. I have a whole thing that I'm asking you to describe. And when it's describing it, it's basically becomes a part of the next prompt that actually works really well. It doesn't always come out perfect, but it increases the chances of it coming very close to the real thing. It's faster than doing a photo shoot. That's for sure.

38:32So you see what I'm doing with this? I've said to it, make the bottle look exactly the same as the image above. It still didn't adhere to that. Make the bottle taller, longer. Then it just messed the whole thing up and made it completely. Now it's way taller. now it's way taller okay so i was like this is my sort of creative mind and this is something you need to think about when you're using these tools is you need to i don't think this is going to be like this forever but for now we have to switch between tools so what i did is i went into mid journey and i cut out the image and i dropped it into i basically took this image which is one of the images that i created cut it out no background and i said i asked chachi to write a prompt for me, create this prompt, put this in here using the Omni reference, which is the subject reference.

39:15And you can see the text is not great. The images look amazing. Like, I think they look great. Like in terms of you're trying to sell a product, this is great, but you can't have that. Now, one option is you take the image into Photoshop, you do a bit of Photoshopping. I'm not trying to get rid of designer's jobs here. I have a lot of friends who are designers, and I know a lot of people get upset about this. But the reality is a designer should know how to do this without having to go into Photoshop. This is the truth. So anyway, I'm going to rush through this. We've got a whole bunch of these images.

39:40So I tried a whole bunch of different weights and whatever, but it wasn't working. It really, it screwed the bottle up a little bit. It started getting a little bit strange here. I changed the background. And then I went with like an ocean background. And then to Iza's point, I took the image back into ChachiPTS and explained in detail what this bottle looks like. That's when I started getting a bit of a breakthrough. Now it started looking a little bit more like the bottle. Eventually it was like, hey, now we're getting somewhere. Again, the text is not perfect. But which tool does pretty well with text?

40:10Chat2PT. So now I've got this image. Go back to Chat2PT. Drop it in and say, make this image more realistic. So this is the image I dropped in. And boom, we've got it. Still got that sepia tone, which I hate. But you get where I'm going with this. So to me, looks like the bottle. But this is how you have to think about using the tools. You have to mix and match. You have to know like, that's better at that. That's faster. That I'm going to, as opposed to just asking the same question and bumping your head, you're in the same context window. Chat2PT doesn't understand that you want something different.

40:42I then dropped in another image. So this was another one created in Midjourney. Asked it to make it look more realistic. And now it's not loading the image. Of course. Yeah, we can see it when it's not zoomed in. So I want to pause for a second. Oh, you know what? Let's go to your next thing because I see what it is. So the last one is turn it into a banner style ad that could be in a magazine, on social media, whatever. And this is without, I just said, stillness has a scent. To me, this looks like something you'd be in a, it's fine in a magazine. It's not loading, but yeah, you get it. So it's used the same image, hasn't changed anything, whilst made it more sepia.

41:16And there you go. There you have something that you can use. Now swap out the product, swap out the background, go play. This is what's available. This is happening right now. This is what you can do and you can do it at scale. If you've got four people running this, the amount of content that you can generate in a day is probably what you could do in a month, about two years ago. I agree. And I'll say two more things. First of all, I'll do a quick recap of everything we talked about because I think it's important for people because we shared a lot of things. First of all, you've got to learn how to use these tools, and you can start very simple.

41:45Start with a very simple prompt, but then what you add in is a structure that includes describing the subject, the fine details of the subject, describing the lighting, describing the fine tones of the lighting. Is it soft lighting? Is it bright lighting? Is it daylight? Is it a studio? What kind of lighting do you have? Describing the background in more and more detail. In Me Journey specifically, as you mentioned, it's important to start. Think about it as the highest weight in the image is going to be what you mentioned first and the weight goes down as you go further into the sentence. So just start with the thing that you care about the most in the image and then work your way in the sentence over there.

42:20And what you'll see, people are sharing cool pictures of you, Ross, as a fox in the chat. You can check that out. So build the prompt with the details of the different components of things you want to do. Provide it as much context as possible. In Chachipiti, it's easier to provide context because you can have an entire conversation. You can give it a story about your brand and your target audience and previous images that actually worked on social media or on an ad campaign. You can give it all of that and it will take all of this into account where on mid-journey, you have to explain that in a single prompt because that's the way mid-journey works.

42:56On the flip side, on the single image, mid-journey provides you a lot more control. You can bring reference images, which you can also do in ChatGPT, but not in the same level of understanding of what a reference is. They now introduced an actual better way to do character reference, which allows you to keep even more consistent characters in mid-journey. So advanced Me Journey usage with actual parameters of Me Journey allow you to get more granular in your control over different components of the image that is not as easy to get to on ChatGPT. But I think the bottom line is, and now I go back for Sokol to what we said in the beginning, you can get to decent results on all these tools today.

43:38it's probably easier to get to an eight out of 10 on chat GPT just because you can explain yourself without being an expert. You can probably get better results as far as image quality and more sophisticated look on mid-journey than you can on either Gemini or chat GPT. And I can tell you myself, and then I'm going to ask Ross the same question. Today, right now, I create, and I switched completely. So until two months ago, I created probably 70 % of my images with mid-journey, another 20 % with ideogram, and then the rest with like flux. Now I'm creating probably 65 to 70 % of the images with chat EPT, just because I can get to a good enough image with chat EPT.

44:23And the rest are between probably flux and mid-journey and ideogram, depending exactly what, and I only go there when I fail getting to the output I want on chat EPT. So I'm curious about kind of like how you see things and how your work is right now yeah so i'm i mean if i have to break it down into percentage wise i'm still majority mid journey because i just enjoy the tool i find myself on the explore page all the time and i draw a lot of inspiration from that um but i would say it's probably 40 percent mid journey and then 20 what's the rest there 35 35 yeah so like no not 35 30 30 i would say that it's majority mid journey and then the other two image effects and ChatGPT.

45:02But again, it depends entirely on what I'm doing. Now, we spoke about it earlier, that JSON structure, which is a coding language structure where you can now use that to build prompts for your images. That changes everything for me and maybe for someone who's more technically minded because I can see in code almost, in a codified format, what the image might look like. I can read that like a sheet music almost. I'm like, okay, cool. I can see what it changed. They changed that. It just makes it so much, just the developer in me just appreciates that. It provides a lot more structure that is very visible versus just a sentence with commas in between.

45:35There's a very clear structure to everything. There's an interesting question in the chat, which is how long do you invest in a quote unquote good image? Well, good is relative, but a relevant image, it depends on what's personally what I'm using it for. If it's for a brand campaign, I spent on one image, I spent three weeks. That was for one key visual where it was backwards and forwards with the client using multiple tools. This was four months ago, so I could probably do it in a week and a half now. That's the reality of it because I know better tools, but that's what it can really take. But once I have an idea and I'm able to ask ChatGPT to spit out different versions of that idea, then it's quite simple.

46:12I can get it with the image I'm looking for within a couple of minutes. Yeah, I agree. And that's, by the way, these two answers are awesome because they serve two different purposes, right? One, they serve, I need a quick something for an ad or a social media. and the other is the client, a client has a very specific output they are looking for and they want it to be perfect and that's just going to take more iterations and trying to understand what they want and showing them different samples and understanding their style and a lot of other stuff. So I think those three weeks or a week and a half are not three weeks or a week and a half of working in Me Journey.

46:43There are three weeks or a week and a half of back and forth with a client, getting their feedback, meeting with them, getting their comments. The actual time probably spent on the tool is probably a few hours. I don't know, two, four. Yeah, yeah, yeah. I forgot that I actually created this as well. So this is just a quick video. So this is the next level of all of it. I mean, out of the scope of what you're doing. Yeah, so again, for those of you listening, we took the ad that was created for social media that has this square around it that says stillness has a scent, and it says Fleece, which is the brand that Ross invented.

47:16So don't go buying this because you won't. And then took the image next to the ocean, kind of like on a rock next to the beach, and used one of the video generation tools, I don't know which one, to just make it live. And it looks awesome. Like it's a slow shot that zooms in and pans to the side. The background is blurry, but you can see the waves kind of like coming in. It's just a beautiful scene. Also, just so you know, I also took this into After Effects and separated the two images from the video and the outline itself. So this is another thing that I think people forget. It's like, you don't have to be fully AI.

47:49You can't. Like the reality is what you understand about the traditional tools that you have used, if you're a designer or any sort of tool that you've used. Whatever gets the work done the fastest and the most effective way. For me, I'm an After Effects, you know, motion graphics artist and animator. So for me, this is logically like, just pull all these in there, I can get way better results. And at half the time as well. So yeah, just a little note on that. There's a great question from Stan on LinkedIn. And I think that's gonna be the last thing because we're running out of time is, can you show what is the prompt that you're using in ChatUPT to have it create a better prompt that you can use in Me Journey?

48:22ah yes that is a great question that is a great question and i hope i have it here somewhere let me see if i can find it yeah and by the way i do a lot of that as well and kind of like cross between creating prompts in one tool to use it in a different tool and then i see the output and then i go back and forth ah here we go can you see my screen yeah cool so i said create an image so i didn't realize but majority has this create an image button so that just saves you from having need to type please create an image of a fragrance bottle the brand's name is fleece in a studio environment and then i created this which i clipped the images out of which is just like a product shot of multiple angles i then said describe this bottle in detail for mid-journey it then said here's a detailed mid-journey prompt for the fleece bottle now it said a studio shot what about it actually described the entire image as opposed to just the bottle i then said just describe the bottle and nothing else and then it said square glass perfume bottle with rounded edges filled with pale blue liquid it has matte black cylindrical cap with a white rectangular label on it with the word fleece this is very important with the word fleece because that is what helped mid-journey like oh okay cool i need to structure whatever's in inverted commas onto this bottle on the label itself fleece in bold black uppercase letters the bottle has clean lines thick walls and a flat base now that is a very descriptive way of taking that into a descriptive prompt that you can use in chatubt along with the image reference which i've used as a style or object reference that's how you get as close to the product shot as possible in mid-journey, take it out of mid-journey, put it into Chattopatik and then create the final product for you.

49:52I love it. I'll say one more thing that again, I've learned that works well when you start to do products, ask it for aspect ratios and ratios between the things. So what's the aspect ratio of the bottle? What's the aspect ratio of the cap? And what's the ratio between the bottle and the cap or whatever it is, this could be again, a dog on the couch, right? Like it gives in the description, the understanding of what this thing is. And it dramatically increases the output of not squishing the stuff that you don't want to be squished. It doesn't guarantee anything, but just increases the chance.

50:19Ross, this was fantastic. I think we covered a lot of stuff. I think we gave a lot of examples to people. And I really appreciate you coming, spending time with us and sharing all your experience with us. If people want to follow you, learn from you, take your course, connect with you, what are the best ways to do that? Yeah, man, please. Thank you. It's been an absolute honor. Thanks. Thanks for having me. I'm very active on LinkedIn. I'm Ross Simmons on LinkedIn. I run a company called Zen Robot, which you can find on LinkedIn as well. And yeah, we are running a four-week masterclass, which is a four-week masterclass in content, I guess, Gen.AI for content creation.

50:53From the beginning, image generation, all the way through to creating images and a final or multiple portfolio pieces that you'll have at the end of that. So yeah, you'll see the links on my pages. But otherwise, you can go to zenrobot.ai and that'll push you to the course page as well. Awesome. Thank you so much. Thanks everybody for joining us live. great chat and cool images of people posting that never happened before. So you inspired people to do stuff. So literally I've been doing this once or twice a week for a very long time. Now, the first time that people are actually posting images.

51:23So that was really cool. Thanks everybody again for joining us. Thank you so much, Ross, for being here and sharing all your amazing experience. And the last thing that I will say, go experiment yourself. The tools are now production ready. That's all that I could say, all of them, regardless of which tool you pick. And especially if you try with all of them, they're production ready. Not for everything. If you need to do a building side printout, then maybe that's not the tool, but you can still use it in the inspiration phases and in the approval phases with the client and then go to upscaling or actual photo shoots and stuff like that.

51:53Awesome stuff. Thanks, everyone. Have a great rest of your day. Thank you.