Create Repeatable AI-Generated Characters in Ideogram

Full video transcript

Have you ever wanted to create a character with generative AI image generation and be able to repeat that character doing different scenes, different poses, different actions, and different facial expressions? If that’s something you’ve ever wanted to do, then pay attention, because I’ve got a special treat for you in teaching you how to do that in Ideogram.

Hi, this is Bruce Fleck, and I want to thank Darby Rollins for inviting me to speak here at the Gen AI University Scale with AI Summit, and I want to teach you how to use Ideogram so that you can get repeatable characters, or at least repeatable styles, in however you are creating your images with generative AI.

Now, I always start with writing my descriptions using ChatGPT or Claude, or one of those generative AI text tools. And so in this case, I just said: I’m writing a blog post for how local SEO is not only viable, it’s the best way for local businesses to be visible. So please describe three different characters with names and visual detail that are business owners of the following types of business: a doctor, an owner of a natural medicine clinic, a dentist, or a physical therapist.

I’m going to use this Dr. Michael Chen here, and I’m just going to grab this description here. Actually, before I do that, I’m going to have it create a before-and-after scene. All right, so now: choose character number two and write two image descriptions for Ideogram. In the first description, Dr. Chen is worried because he is not getting enough new clients. In the second image, he is happy and confident, as he is getting plenty of new clients using local SEO techniques. For the description, output it in a photorealistic style, using the structure style, subject, setting, and optionally text overlay.

Okay, good. All right, so we’re going to go over here to Ideogram. This is what the home screen looks like. And then you click over here on “Creations.” So you click up here at the top and paste in your prompt. And for now I’m going to get rid of this text overlay, because I don’t want that confusing what we’re teaching here. So we’ve got this first one of Dr. Michael Chen.

You have the choice of making it public versus private. If you do private, then it costs more. And you can choose any aspect ratio you like, but for a blog I would use this 3:2. You can choose how many images you want to produce. I’m going to produce three — actually, I’m going to do four. If you would like it to use magic prompt, where it will enhance the prompt, you can do that. You can turn it on or off. If you have a previous character, this is where you will choose that, or if you have a previous style, you can choose that. I’m going to exit that out because I’m just going to let it run with what it’s got here the first time.

All right, so I’m going to generate that. I’m going to use four credits this first time because I want to get a good character. I’ll pause briefly while this is generating. Okay, so I’ve got four pictures of dentist Dr. Michael Chen. These two look pretty good, that’s also pretty good, I don’t like him, so I think I’ll use this one — and he’s got the “Chen Dental,” that was a nice touch that it did, Dr. Michael Chen — that’s also good, but I like this image here.

All right, what I’m going to do is go back and grab the second prompt again. I’m not going to do the text, just grab the second prompt. And then what you have to do is go back up here and pull all the old stuff out, paste the new stuff in. And then I can choose a character. Now, I don’t have one chosen already, so I can either click the plus icon and click there, or I can click down in here and go on the three dots and choose this right here, and it will automatically add him as a new character.

All right, now I’m not going to make four of these, I’m going to just make two of them, and we’ll see if we can find a happy Dr. Chen here with this next generation of images. And I think you’ll see that it’s very similar. I’m going to go ahead and delete this one. Okay, so this is the unhappy Dr. Chen — okay, that’s him there, pretty close, that looks very good. So I like this, I like this one, and I like this one. Okay, so it did a pretty good job of making this character and this character look very similar.

Now, if I didn’t want to make the exact same character but wanted to just follow the style of this, then I turn this off, and instead I go to style and I click on reference. Oh, I can upload my own as a reference, or again I can click on the three dots and go on style, and I’m going to generate, and this time it’s not going to be the same character, but it will be a very similar style to this image right here. So you can see that these are different characters but in the same kind of imagery style.

Now that worked very well for me when I was using illustrative styles. So here’s an example: I had this style here, and in fact I even had — I had this style here and I liked this style very much, and so I wanted a different image with a woman, and so then I got this style here that I liked, and then I wanted to repeat that style again, and I got this image here.

And one of the nice things about Ideogram is that it’s got this Canvas feature. What I was able to do was click in here and open in Canvas, and I was able to get rid of this starry night background — I just got all that out, erased it out, and was able to put in text on top of it. So if I go back, just like our previous ones with Dr. Chen here, I could do the same thing.

If it was — I think this is the one I liked — and if I want to write some text over here, I can open in Canvas and I can do a variety of things. If I wanted to take these people out of the image, I could remix or do magic fill, or I could extend it if I wanted to make it wider or taller, any of that stuff. But I’m just going to show you the text feature right now. I don’t like that style, so I’m going to choose a standard Montserrat, 64, bold, and choose a black text, and I will click over here — find more patients, local SEO — that color is not working for me, so what if I just went to white, or maybe I went to a shade of yellow. Okay, and there we go. And then now I could download this and use it as a social media post or with my blog post or whatever I wanted to do there.

Yeah, you’ve got text editing features in here — again, you can remix and extend, use magic fill for all those kinds of things, and it gives you lots of options here in Ideogram. So I like it a whole lot.

There’s one last thing I’d like to show you. I had uploaded a previous image of myself that was generated with another image generation platform. I trained it on various pictures of myself and got a good picture, and it became this character reference here that I had uploaded. And what I did was then I gave it a prompt asking it to show me holding my new book. Now, what I was trying to do was to get it to show my exact book cover, but that didn’t work. But you’ll see that I got some really good images of me that really look like me, and I did a great job of keeping my character consistent.

And so I feel very confident now that if I wanted to have pictures of me doing anything — whether that’s riding a bike, or speaking in front of a crowd, or meeting with a client one-on-one — that I could use this image and it would do a pretty good job of that. I really like Ideogram, what you can do with characters, character references, and style references, and the Canvas. So it just gives you a lot of flexibility to do what you want to create the kind of images you want with generative AI.

So with that, I hope you learned something valuable, and I want to thank Darby Rollins again for everything he does at Gen AI University, and enjoy the rest of the summit. Thank you, bye.