I’ve been messing around with various generative AI applications, just like many of you. I’m trying to figure out how I can use them in my business and also get really good at using various AI tools. Since I create online courses, I’ve been working on making my own illustrations that work well.
In a previous post, I shared some tips based on things I learned using AI. I’m using Midjourney for my examples, but there are plenty of tools out there for initial experimentation.
Here are some tips I’ve picked up and how to make those free illustrated characters that I share on LinkedIn.
It starts with instructions
The secret to getting the right pictures is based on motivation. The challenge with AI is that you never know what you’re going to get and once you get something you really like, you might not get it again.
- I start simple and don’t try to overdo it or be too descriptive in the beginning. And then I work from there.
- I also look at what images people are sharing in the AI community and when I find something I like, I try their queries to see what I get.
- I generally use character, plot, and then style for instructions. Depending on what I need, I may add a location or something to provide more visual context.
Character
There are a lot of biases built into the language models of artificial intelligence. For example, if I type businessman, 99% of the time it will be a white businessman in a suit. Why white? Why a suit? This is something you have to take into account when creating characters.
Regardless of prejudice, I start with something generic like a businessman. That usually gives me a white guy in a suit. But that’s fine for a start. Once I get the look right, I change the descriptor. A few tips for better initial signs:
- Instead of just “businessman”, use age to see what you get. I find 40 year old is a good addition.
- It doesn’t matter what you get, the main point is to see if you can come up with something consistent.
- Now there are many good videos on how to write instructions. The more you practice and play, the better you get.
- At some point, I say it’s good and move on. My goal is to find an easy way to get a good enough image. Playing with the AI for hours (or days…whatever will happen) is fun, but not something I can afford to do for long.
As you can see from the images below, you get pretty decent initial images to start with. I can work with these.
Action
By describing the action, you get the pose you want and can use. I keep it simple.
- I like to throw in “gesturing” as part of the query. It allows me to have more different hand positions.
- Add an emotion or state like “confused” or “angry”.
- I also change the pose and can use “arms crossed” or “hands on hips”.
- Play around and see what you get.
You will notice how the images change based on a few simple words.
As you browse the images above, you’ll notice that the styles change quite a bit. You will also notice that you may not get the pose you expect. For example, “hands on hips” includes crossed arms. “Crossed arms” are not full body poses. It’s something you have to live with.
As you practice and learn how to write instructions, you’ll be able to design more consistent styles.
Style
As mentioned above, achieving a consistent style can be difficult. A lot of it will come down to how you write the instructions and the feel you get for them as you practice.
- For the illustrated characters, I want a similar style and no background so they are easy to clean.
- My default quick style is “flat vector illustration” because that works well for a lot of eLearning course designs.
- It also works consistently most of the time. Even if the characters illustrated are not the same, they are often close enough to be used together.
- I play with unique style descriptions. For example, I will specify “in style [add descriptor].”
As you review the images above, you’ll notice that adding a style description to “flat vector illustration” starts to provide a more specific style that approaches consistency.
Additional style tip: get rid of the background
The AI image is just a starting point. Often you may need to edit hands, faces, etc. One of the main things is getting a transparent background so I can place the image where I want.
In order to get the cleanest possible background, I add the query: “white background.” It works 90% of the time. The image below is the same “hands on hips” query as above with the addition of “white background” in the query.
Overcoming Bias Using Artificial Intelligence
As I mentioned before, there is a built-in bias that can include things like gender and race, but also clothing style. You can mitigate some of this by changing the descriptor. Once I get an image style that I like, I’ll start adding more descriptions to give more variety to the characters. In no time I have a diverse set of characters to build from.
One thing to consider is your own bias. For example, I may search for a particular image and then assume that it works because of how “I see” that character and what it is supposed to represent.
These are some basic tips to get started creating your own characters using artificial intelligence. In the next post, I’ll share my secret sauce and what I do to speed things up and how I edit characters.
Don’t forget to download the characters I’ve already added. I’ll be adding more, so keep your eyes peeled.
Events
Free e-learning resources
