I’ve been messing round with completely different generative AI apps, similar to lots of you. I’m attempting to determine how I can use them in my job and likewise get actually good at utilizing the assorted AI instruments. As a result of I create on-line programs, I’ve been engaged on making my very own illustrations that work effectively.
In a earlier publish, I shared some ideas primarily based on issues I’ve studying utilizing AI. I’m utilizing Midjourney for my examples, however there are loads of instruments on the market on your preliminary experimentation.
Listed below are some ideas I’ve picked up and the way I make these free illustrated characters that I share on LinkedIn.
It Begins with Prompts
The key to getting the suitable pictures is predicated on the prompting. The problem with AI is that you simply by no means know what you’ll get and once you get one thing that you simply actually like, chances are you’ll not get it once more.
- I begin easy and don’t attempt to overstate or be too descriptive initially. After which I work from there.
- I additionally have a look at what pictures individuals share within the AI neighborhood and once I discover one thing I like, I attempt their prompts to see what I get.
- For prompts, I principally use a personality, an motion, after which the model. Relying on what I want, I’ll add a location or one thing to offer extra visible context.
The Character
There’s loads of bias constructed into the AI language fashions. For instance, if I sort in businessman, 99% of the time it will likely be a white businessman in a go well with. Why white? Why a go well with? That’s one thing you want to account for as you create characters.
Whatever the bias, I begin with one thing generic like businessman. That often offers me a white man in a go well with. However that’s OK to start out. As soon as I get the suitable look, I modify the descriptor. Some tricks to get higher preliminary characters:
- As an alternative of simply “businessman,” use an age to see what you get. I discover 40-year-old is an effective addition.
- It doesn’t matter what you get, the primary level is to see if you happen to can give you one thing constant.
- There are loads of good movies now on easy methods to write prompts. The extra you apply and mess around, the higher you get.
- Sooner or later, I name it good and transfer on. My aim is to discover a easy approach to get a ok picture. Taking part in round in AI for hours (or days…which is able to occur) is enjoyable, however not one thing I can afford to do for lengthy.
As you possibly can see within the pictures beneath, you get some fairly respectable preliminary pictures to start out. I can work with these.
The Motion
Describing the motion is the place you get the pose you need and may use. I preserve it easy.
- I wish to insert “gesturing” as a part of the immediate. It offers me extra different arm positions.
- Add an emotion or state-of-being comparable to “confused” or “indignant.”
- I additionally change the pose and will use “arms crossed” or “fingers on hips.”
- Mess around and see what you get.
You’ll discover how the pictures change primarily based on just a few easy phrases.
As you evaluate the pictures above, you’ll discover that the types change fairly a bit. You’ll additionally discover that you could be not get the pose you count on. For instance, “fingers on hips” consists of arms crossed. “Arms crossed” aren’t full physique poses. That’s one thing you must stay with.
As you apply and get to know easy methods to write prompts it is possible for you to to design extra constant types.
The Type
As talked about above, getting a constant model could be powerful. A whole lot of it would come all the way down to the way you write prompts and the texture you get for them as you apply.
- For illustrated characters, I need a comparable model and no background so it’s straightforward to wash up.
- My default immediate model is “flat vector illustration” as a result of that works effectively for lots of e-learning course design.
- It additionally works constantly more often than not. And even when the illustrated characters aren’t the identical, they usually are shut sufficient to make use of collectively.
- I do mess around with distinctive model descriptions. For instance, I’ll state “within the model of [add descriptor].”
As you evaluate the pictures above, you’ll discover that including a method description to the “flat vector illustration” begins to offer a extra particular model that will get nearer to consistency.
Bonus Type Tip: Getting Rid of the Background
The AI picture is simply the place to begin. Usually you might have to do some edits to fingers, faces, and so on. One of many major issues is getting a clear background in order that I can place the picture the place I would like it.
To get as clear a background as attainable, I add the immediate: “white background.” That works 90% of the time. The picture beneath is similar “fingers on hips” immediate as above with the addition of “white background” within the immediate.
Overcoming Bias with AI
As I famous above, there’s built-in bias which may embody issues like gender and race, but in addition model of garments. You possibly can mitigate a few of that by altering the descriptor. As soon as I get a picture model I like, I’ll begin to add extra descriptions to get extra various characters. Very quickly in any respect, I’ve a various set of characters from which to construct.
One factor to contemplate is your personal bias. For instance, I’ll search for a particular picture after which assume that it really works due to how “I see” that character and what it ought to characterize.
These are among the fundamental tricks to get began creating your personal characters with AI. In a comply with up publish, I’ll share my secret sauce and what I do to hurry issues up and the way I edit the characters.
Don’t neglect to obtain the characters I’ve already added. I’ll be including extra, so preserve your eyes peeled.