https://www.loom.com/share/8e000eb60c184374a2b091feeb5b0364?sid=5a6b1b3e-f6dc-4b32-9e72-b714d0e0f334
https://chatgpt.com/ - Dall-E
00:00 Hey guys, in this video we're going to be looking at how to create standout on-brand images using AI. Now the, the limitation with some of these AIs still is it's difficult to produce consistent outputs, there's variability. 00:17 But if you can get your prompt nailed and use the same prompt structure every time you create an image it becomes much, much easier, yeah. 00:24 So, it's a bit of effort to get it working to begin with, but then once you've got your prompt nailed it becomes really easy. 00:31 So, what we'll do is we'll start with a, with a reference image. Yeah, so I've just found some of this doodle art style, um, stuff, um, that Notion, Notion seems to have the, the lion's share of the brand ownership over right now. 00:44 Um, if I choose this, um, so I've just got an image example there, and then once it's uploaded, this is mid-journey by the way, it starts at I think ten, ten dollars a month, something like that, um, so it's peanuts for how powerful it is, um, doesn't seem to want to upload my image, let's try again, 01:04 shall we? I've uploaded the image, and when you've uploaded your image, you can click on this little eye, and what it will do is it will analyse that image, to give you descriptors, known artists, and these are good to copy to bring into a prompt. 01:29 So, uhm, let's just, I'll just get up a new notepad. So, So, uhm, let's do that. 2D flat design style. 01:58 And then, simple line drawing. And then, let's do no shading with simple shapes and bold lines. 02:16 Okay, now what you want to do then is you want to, we've got doodle style in there. So let's do a simple line art with, you don't want the prompt to be too long because sometimes it can detract from the accuracy of the scene you try and get it to do. 02:44 There's probably, like, AI image gurus that are probably saying that I'm doing this all wrong, but this is what worked for me. 02:53 So what I'll then do is I'll break this down into a simpler prompt. So we can say a simple vector illustration. 03:05 Flat 2D design style, black and yellow color palette, flat 2D design style line, line drawing, uhm, no shading, with simple shapes, and bold lines, uhm, no shading with, So let's do, cartoon doodle style, cartoon doodle style, no shading, with simple shapes, and bold lines. 04:04 So simple vector illustration, flat 2D design style, uhm, line drawing, uhm, and then let's do black and yellow color palette, cartoon doodle style, no shading, with simple shapes, and bold, okay, fine. 04:20 So that then becomes the prompt that we use to create, uhm, all of our images after we've asked, uhm, what we want it to do. 04:33 So what I tend to do is this. So if we just imagine this, okay, a simple vector illustration, of, enter, subject matter, and then, black and yellow colour, there we go, cool. 04:58 So we've now got a prompt template with something to enter. So I can now go in and create a new Chrome. 05:08 Let's just go back. There we go. And now I can copy that template. So a simple vector illustration of, uhm, someone riding a bicycle. 05:30 There we go. So now mid-journey I'll copy Have a think about that, uhm, and then hopefully it will be somewhat close to the image that we uploaded. 05:39 If it's still a bit far, then you can add complexity to the prompt. You can try switching out some of the descriptors with some of the other elements, uh, from the original image. 05:47 Uhm, but we're not, we're not too far off. Yeah. What I might want to emphasize is, uhm, I want it to be a white background. 05:56 Missed that out, which is fine. So, so the AI, uhm, is giving me a combination of yellow backgrounds and, uhm, white backgrounds. 06:05 But again, it's just a case of iterating on that prompt until you get the one right. But what's good about Midjourney is the previous images that you've produced, you can just click on the text and it will bring it back into the prompt. 06:15 And that's exactly what I do. Yeah. So for me, I'm more sort of outrageous 80s synth wave. You know, that's, that's my prompt and then I'll just do a stunning black and white outrageous synth illustration of, and then enter subject matter with then those instructions on how to follow up So that's, that's 06:32 easy enough. Um, if you don't want to pay for Midjourney, you can do it in ChatGPT. Um, DALI isn't quite as advanced as Midjourney. 06:43 So I tend to find that it doesn't create. Images that are quite as, as natural. Um, but, um, it is quite quick. 06:50 And again, it comes as part of, uh, ChatGPT. So here, what I've done is I've uploaded the image and I've just said, I want to create an image prompt that I can use to create consistent repeatable images in the style. 07:00 So, um, what I want to do before creating this prompt, um, is just start with an image analysis. So we've got style, color, character design, composition. 07:09 And so on and so forth. Um, so what I'll now say is, right, okay, right. Can we condense this analysis into a two-sentence prompt on brand images? 07:33 Okay, so hopefully this should do the trick Yeah, see the only problem I've got with this is the scene should abstractly represent themes like creativity, project, teamwork I don't want that. 07:55 I want it to be a blank template So I want this to be a blank template So can we update this time to be slightly more descriptive but excluding this and then to the bit that you don't want, there we go So, when I now create a new, uhm, prompt, there you go, so we can say image, but in reality you can 08:56 just say create an image, so I'm just going to copy this, so, uhm, okay, so create minimalistic clean illustration of And then let's break that up as well. 09:14 Someone riding a bicycle. Simplified form, there we go, cool. Uh, there we go. The one thing that is quite nice about ChatGPT is you can continue the conversation. 09:40 It's a little bit difficult to continue the conversation in mid-journey because it's not a chat. Uhm, okay, fine, well, that's not really anywhere near what I wanted it to be. 09:48 So you can, you can see that, yes, you know, uhm, it might be okay for ChatGPT, but we might need to go through quite a lot of iterations to get it to the point where it actually looks like what we want to. If you want, you can always upload the image, yeah, so you can have a reference image. So if I 10:17 do this again, so if I just create a new chat, uhm, so let's do image example there, and we can just say, create to me. 10:21 an image, uhm, of someone riding a bicycle in the same format as the attached image, yeah. But the only issue I've got with doing it this way is it's not, it's not really your own if you're using a reference image every time. 10:45 Thank you. Yeah, so you might want that style, but you might want to prompt to, to get a similar style even though it's not identical. 10:51 Uhm, so, either way, that's, that's an option for you. We'll see what it comes up with. But reference images are probably going to be more accurate with ChatGPT than trying to create a prompt. 11:02 Okay, so it's a, it's a bit better, uhm, but not great. 11:19 Uh, what is nice about ChatGPT as well is, uhm, you can edit. So I can say stuff like remove. 11:29 And that, that is quite, quite a nice feature to be able to manipulate as you go through. You, you can't do that in mid-journey. 11:42 Alright, uhm. Let's just see it produce a final output and we'll leave it