Transcription
This is Maya. Finally here in Rome, Italy. Look at this amazing Trevy fountain behind me. It is absolutely beautiful.
She doesn't exist. No photographer, studio, location, or model agency needed. Just open art and a process you're going to know by the end of this series. Oh, him too. But we'll get there.
This is the series. Four stages. Each module builds on the last. What you create today feeds directly into the next video and the one after that. By the end of this series, you'll have a fully consistent [music] trained AI influencer you can generate new content with on demand. Any environment, outfit, any scenario without starting from scratch every time.
Module one is where it all begins. Build it. We're choosing the right models, learning how to prompt, and building your character's main look. Everything we make today has one purpose, setting up the foundation for what comes next. Let's dive right into it.
First, model selection, and specifically the difference between two types of models you'll see in Open Art. Multimodal is what we need for building a character because at some point you'll have a face you like and you'll want to change the lighting or the outfit, the expression without rebuilding from scratch. Multimodal lets you do that. Generationon models doesn't.
These are the models that I suggest that you use. either Nano Banano 2 or Pro, Seed Dream 4.5, Flux 2 Max or Flux Klein 9B, and Quen image 2. Any model that is labeled reference is multimodal.
Here's the exact same prompt run through four models. Same words, four different interpretations. Notice the difference in the skin texture, lighting, and overall feel. None of these are wrong. They're just different directions. Your job is to figure out which model fits your character style best. Everything I'll be showing you today, we will be using Nano Banana 2.
Now, let's talk about prompting. One thing to clear up before we go any further, as long as you follow this basic format, there really is no wrong or right. It comes down to what feels natural to you. This is an example of a very simple prompt that I would write. That's how I describe a shot to a photographer. It feels like a direction, not a list.
The latest models understand natural language. You're not writing code. You're describing what you want clearly and specifically. And the word specifically is the whole game. AI models are like toddlers in a sense. Not because it's not smart. it is. But because it understands the language perfectly and still needs very specific direction to get to its destination. You could tell someone to turn left somewhere or you can say turn left at the third traffic light after the bridge. Now, that's very specific direction. Don't just be detailed. Give context. Be specific. Every part of your prompt is an opportunity to remove a guess the model would otherwise make on its own.
One more thing, filler words. Very, really, a bit, kind of. Leave them out. The model doesn't need encouragement. It needs direction. And a quick note on using chat GPT or Claude to write your prompts. Hold off for now. Those tools tend to overprompt. When something goes wrong with a dense generated prompt, you won't be able to know what to fix. Prompt manually first and learn how the model responds to your own language. We'll revisit AI prompt tools as a bonus in a later module once the foundation is solid.
One decision to make before you generate anything for your own characters, what is the visual style? photorealistic, 3D rendered, animate, 2D. This matters more than people realize because everything you generate today becomes a reference asset for the next module. If your images are visually inconsistent, the foundation is inconsistent. Nail the style now. Everything becomes easier from here. And in terms of prompting, I like to put it at the start of the prompt. cinematic photo of 3D CGI render of anime manga style of you get my point.
This is the most important generation in the entire module. When you're describing your subject, go deep. Don't just say a woman in her late 20s. Describe everything that makes your character unique. hairstyle, hair color, nationality, facial features, skin tone, body type, any distinctive detail that makes your character unique.
Two things worth knowing that most people don't think about. First, mixing nationalities is one of the most effective ways to get interesting distinctive looks. Asian European gives you something different from East Asian or European alone. The model interprets these blends in genuinely unexpected ways, and you'll often land on a look you never could have planned. Second, giving your character a name in the prompt will subtly influence the output. Names carry cultural associations, aesthetics, connotations, a certain feel. Maya generates differently from Ingrid or Yuna, Valentina. Try a few and you'll see it. Is this scientifically precise? Not really. Does it work enough that I keep doing it?
The second approach to building your character's face is combining multiple reference images to arrive at a look that's entirely your own. For this example, I'm using reference images of the Blackpink members, Lisa, Jenny, Ros, and Jizu. With the four references loaded, I'm keeping the prompt intentionally minimal because I want the references to do the creative work here. I'm just giving the model a style direction. This is the technique for when you have a visual feeling but can't find the words for it. Find references that carry that feeling, combine them, and let the model translate the inspiration into something original.
Now, everything I just showed you applies to any character you want to create. Any age, style, or vibe. I also didn't forget about the dudes. I just made you wait a little. Same process, completely different energy. You're welcome.
That, my friends, is module one, building a solid foundation on developing your character. In module two, we'll focus on refining your character, adding assets, outfits, anything else associated with your character. So, what are you waiting for? Head on over to module 2 and I'll meet you.