Transcription
Hi, so I got asked this question: "Hi Helena, could you explain to me how you create images with AI?"
I thought, what better way to answer that question than by creating a video outlining step by step how I do it?
There are actually two ways to create amazing images. One is a general method where you prompt with words, and the other is where you prompt with words but also use a different tool to create a token that will feature your face inside whatever image you're creating. I will show you today how to do that.
So, let's go!
The first step to your vision board is actually to have your inner images. If you already have them, that's great; we can jump right into it. If you don't, I would like to recommend checking out my video about how to do the inner work and find images through some templates that I've created for you for free, which you can download from my Substack.
So, that being said, let's go straight into how to write prompts so AI does what we want—oh, I would say mostly what we want.
We have opened a tool—in this case, the paid version of ChatGPT—but you can also use the free version of ChatGPT. I will show you a tool later where, when you create an account, you get a certain amount of credits, and you can create pretty cool images without one. Never mind, later.
Let's do it now! I can recommend a tool called Imagine.art, where you get 50 credits and use about 10 credits per image. So, you can create five amazing images every day by using this tool.
This is one that I created with Imagine.art. So, let's say you are dreaming of a house. Rather than saying, "Show me a house with two stories," make it really specific. Say, "I would love a cozy cottage at the Pacific coast that has bay windows and an ocean view."
I would show you the difference. That's a pretty nice house! Now, I don't know if this is what you're looking for. Again, my style is more cottage style—maybe some rounded rather than the kind of modern version—but it's beautiful.
This might be exactly what you're looking for. If not, let's see what happens. Oh wow, I like this one! Dreamy! Although, I have to say, since living in the United States, I know these types of windows and houses are usually not insulated, but it's pretty. It's very, very pretty. Ooh, it's pretty!
Okay, you see already this is the emotional impact the right type of image can have on you and your vision. I'm totally emotionally hooked right now; I can't even take my eyes off this one because that's a dream for me.
Okay, I'm probably going to download this and put it on my vision board with some little tweaks or something because—nice!
Okay, so tip number two is to use smooth words. Let's say your mornings have been a little bit hectic last year, so you would like for them to be a little bit more calm—kind of like having time in the morning to enjoy your coffee.
So, let's say that, like one of my friends, you have a thing for craft coffees in the morning. "Give me an image of a serene and gorgeous morning coffee cup next to a cozy armchair with a colorful pattern and a cat napping and waiting for cuddles in this peaceful atmosphere."
Let's see what happens. Ooh baby! Okay, we have run into a little problem. I don't like that the coffee cup is on the floor, so we are going to improve the prompt, which is sometimes really important.
On a small table—let's see how that improves. We want to be specific. Remember, I love this! I love this! I've been dreaming of having a cat, but because we are renting and we are sometimes in Europe for extended periods of time, we don't have a cat.
Oh my God! Oh my God, look at him! It's a ginger cat too! Oh, okay, so this is much more emotionally wrecking than I expected, to be honest.
Okay, let's say something else. Let's say you're dreaming of a kitchen remodel for your parents. Okay, we have to be specific.
So, your parents are dreaming of having you guys over, and they would love to have a cozy dining area with lots of light. Now, of course, I don't know what this kitchen looks like in general, so I could be much more specific if I had a photo in mind, but let's see what happens.
Okay, not bad! Very stylish! Actually, I do not like the chairs. I do not like them at all. I can already see that one chair is going to be a pain in the back, but it's pretty.
So, we can either now tell ChatGPT in one or two sentences how we would like the prompt to be improved, or we can take the initial prompt and change the verbal prompt. That's what I like to do quite often because ChatGPT tends to add things that I don't want it to add.
Because your parents want to have you over, or your friends, or whomever you're designing this kitchen for, they want to sit with you for a long time, so the chairs have to be comfortable.
Okay, let's see what happens now. That's too muted. Mmm, that's also too full. I prefer the first version.
Now, if I don't want to do excessive prompting—because sometimes it can take up to 15 images for me to kind of get what I want—you can either take this one out or just prompt until you have what you want.
You can also give feedback, and now it could get better or worse, but sometimes it's a question of having patience to get exactly the images we want.
That's not bad at all!
Yeah, so let's say you wanted to create one of your dreams that contains you in it. I have a dream where I'm wearing a dress that is feeding hummingbirds by concealing nectar tubes behind large flowers that would be sewn and incorporated into the dress.
I have created a more generic version of this that has a lady wearing the dress, and I've created the dress as a version of me wearing the dress. Let's see what happens.
So, I accidentally already prompted this. This is a dress; it's not quite what I was looking for. Design a realistic green dress. Give it large, vibrant, three-dimensional flowers. Incorporate it into the fabric. There are hummingbirds swirling around—obviously, I'm basically a walking feeder.
So, let's see what happens. Not my likeness in it. There will be one with my likeness in it. It's okay; I'm kind of missing a little bit of the environment around it.
I like that it has sleeves. Did we prompt sleeves? Were we specific enough?
Okay, I think it remembers me from prompting this before. We want to see the whole thing in a rainforest environment. Ooh wow, that's very dramatic!
Yeah, not bad! Her face is not quite what I was looking for, but we're getting in the right direction.
Let me show you how this looks like if it was prompted with my likeness and unspecific hummingbird flowers. So, this is me designing a realistic dress. Helena wears it, so there's the connection to the token that I've created with my photographs as a basis.
And then this is what comes out of it. So, the goal is still to create a realistic dress. In this case, I had token Helena wear it, and it's beautiful, but it doesn't have the three-dimensional flowers that I'm looking for.
This is what it looks like when token Helena is wearing the specific dress that I've designed. So, there are two versions of this. I think they're both pretty specific; they both convey the idea.
They're not set in a rainforest environment because the original model that I've created this from has me being in the garden with roses. So, the model is kind of convoluted with different imagery.
I could fix that by creating more neutral models and then adding rainforest images into the folder that contains my face.
We can create both terrible images and amazing images with AI. Just be prepared to have an outcome go either way. It can be sometimes a little bit nightmarish, and it can also be extremely positively surprising.
AI can iterate in our ideas, and we can then continue to develop more of something that is more co-created than just our own idea.
I would love to see you in my next video, where I will show you how to do prompting with your own likeness in it.
I would also love it if you want to comment below about one of your dreams so I can incorporate it into my next video and visualize it.
Bye!