How to use AI to make game illustrations

When we create a game, we need to make game illustrations. Sometimes, we’re stuck because of the lack of images.

Here are the two traditional ways to generate game illustrations:

1️⃣ Shall we hire a graphic designer?

They do a great job that’s worth the price. But if you start a new project, you might not be ready to pay several hundred of euros for each game.

2️⃣ Or shall we rely on stock images?

There’s a picture of everything. However, your players might be familiar with most of them. This gives a less unique identity for your game. Besides, they usually are expensive (around 100 USD)

There’s a new third way, and you’re going to love it:

3️⃣ Using artificial intelligence (AI) to make game illustrations.

So, how do we do this?

There are many existing platforms that allow you to do this. But let’s focus on two of them:

Stable Diffusion

• Quality: ⭐⭐⭐
• Access: From your browser
• Commercial usage
• Pricing: Free

Midjourney

• Quality: ⭐⭐⭐⭐⭐
• Access: Discord
• Commercial usage (if not free trial)
• Pricing: Subscription-based. From 12 USD per month

Let’s see how we can bring new illustrations for our game the Dark Manor. Indeed, I think that this games lacks of game illustrations that would bring a better atmosphere. If you’re curious, you can play it from this page for free.

Let’s start using Stable Diffusion and Midjourney.

In both cases, they ask to use a prompt. But, what’s this? A prompt is a sentence that tells your AI what to generate. No matter how’s the AI, you will need to give an accurate prompt.

At the end of the article, I will provide a few tips on how to make a good prompt.

So now, we’re going to make illustrations for our game! In this article, we will need the following:

1. A picture of the dark manor outside, surrounded by a forest (easy) 🏚️
2. A bathroom in a dark manor (intermediate) 🛁
3. A living room with a big TV in the middle (hard) 📺

Let’s go!

A picture of the dark manor outside, surrounded by a forest (easy) 🏚️

This first task is easy. Manors are often surrounded by a forest.

Let’s type the following prompt in Stable Diffusion:

After only a minute, the AI comes up with 4 pictures to offer.

I like the top left one because it shows a small house, such as our dark manor. I can simply right-click and save it. Stable Diffusion allows it, as there’s a free commercial license on every generated picture.

Great!

Let’s type the same prompt with Midjourney.

Here is the result.

Hum… The result is better. I could use them as I’m personally on a paid plan. But if you’re still on the free trial, then you might not use them for your games.

A bathroom in a dark manor (intermediate) 🛁

This one will be tough. Who thinks of a bathroom in a creepy dark manor? I can’t find it as stock pictures. Indeed, on the number 1 database of stock pictures, you can see here that there are no satisfying results..

Let’s try it with AI!

Now let’s ask for “bathroom from a dark manor, style from Amnesia video game” on Stable Diffusion. And I chose one that stood out from others:

The toilet is not perfect, but the overall picture is excellent!

Notice that I ask a style from “Amnesia“. A good prompt always references a style/artist/movie/video game, etc. Thus, a good culture is required to handle an AI generator! All my hours spent on Netflix or on my Nintendo Switch/Wii/3DS/DS/Gamecube will actually become useful.

Okay, now let’s try with Midjourney.

Again, the result is better. The picture is clearer. The ambiance is closer to want I wish. But it’s not free.

A living room with a big TV in the middle (hard) 📺

Do you like challenges? I do. That’s something important for gamification.

So here, the challenge is to combine 2 things that would not have any relation. And yet, my game does have such a room: A living room with a TV, and a riddle from it.

Let’s try with this prompt, hoping that a Samsung Smart TV appears well in the middle of our room…

Hum… I do not feel satisfied with the previous results. Do you see any TV screens? The bottom left has. But the background is a bit chaotic.

Let’s try again! AI has plenty of time on its hands to help us out.

Okay not bad! I like the top left. With a bit of Photoshop, I would make a screen related to the riddle. And I would move a bit the feet of the TV.

Any AI-generated will always need a bit of work with photo editing software, and that’s normal. It especially happens if your picture displays hands: The AI still doesn’t know how to properly generate hands (and that’s a big topic, as this article mentions it!).

Okay, now let’s use the same prompt on Midjourney.

With Midjourney, the results are satisfying, even from the first attempt.

Same story here, I would simply photo-edit the screen to add something related to the riddle in the game.

Midjourney is overall better than Stable Diffusion to make game illustrations.

But Stable Diffusion is currently free, and its pictures are free for commercial use.

On the opposite, you will need to pay for Midjourney to have the right to commercially use it.

The pricing is based on the GPU (graphic processing unit) usage time of the provider. And in my own estimation, you will pay 10 USD + VAT for… 200 pictures. If I use the AI 3 times to refine your pictures (which I usually do), it means that each picture will cost me 5 cents + VAT.

That’s much cheaper than a stock picture of 100 USD!

To put it in a nutshell, I think there’s a revolution when it comes to make game illustrations.

However, you’ll be only satisfied if you provide the AI with the right prompt. And that’s where skill is still needed.

If you want to be considered skilled, you’ll need to master prompt generation according to your needs.

So, to give a few elements on how to make a good prompt, here are 5 non-exhaustive tips:

1. Be careful with homonyms (e.g. rock, to book, a bat)
2. Expect strange results from the AI if you give words with open interpretation (e.g. fantastic, smart, brilliant)
3. Do you want to anticipate what one word would look like? Google it first. Google image will give you a global view of how the AI interprets it.
4. If you use Midjourney, you can add an URL of a picture in your prompt for more accuracy (here is how to get an URL from a picture).
5. Use your culture to get a style that satisfies you. Van Gogh? Fire Emblem? 300? Each of them has a specific style that could bring your game better.