A step-by-step breakdown of how we are using Midjourney’s AI image generation tools to create concept art for our web3 game, EON Guardians.
Today I’m going to dive into AI Generation as it applies to concept art. We will be using it in the earliest phase of ideation to find the first visuals that help establish the world for players. Note: this is not a “getting started” guide for Midjourney, but a deep dive into a practical use case as it applies to game development or film pre-production.
Let’s quickly explain how Midjourney works: first, you feed it a prompt which generates 4 variations of image and 2 job options for each “[U]pscale” and “[V]ariations”. The AI will generate images based on the prompt and those parameters.
With that out of the way, let’s get into the process.
Step 1 — It starts with the idea that we are trying to realize in visual imagery. For myself, I start by just listing out factors that I know are important to the concept, for example:
- What is the setting?
- Where are we geographically? How do we show that?
- Is there magic or technology?
- Dystopian? Utopian?
- Who lives here? Are they alive or dead?
With those thoughts in mind, let’s go through an example. First Midjourney prompt attempt:
“post apocalyptic town in fjord, anime”
Step 2 — I’ll be honest, I got lucky. I didn’t think this prompt would work out at all. The concept I wanted to communicate was a small but dense civilization in which a mix of architectural elements from different time periods are juxtaposed with a picturesque landscape near water. #4 was giving me a handful of those elements, so down the rabbit hole we go, V4!
Step 3 — What I was really looking for was more prominent geographical formations in the image. While the mishmash stack of houses was nice, I want to show that this is a town that started small but was built up organically into the geography. #3 gives me a new cliff, but I’m curious to see what else I can get, V3!
Step 4 — This is where we show that we know when to stop. Our new variations are either too apocalyptic or too much mountain. So let’s back it on up, we still liked #3 of the previous batch, so let’s U3 (upscale) it!
Step 5 — Ok, I was surprised to find an image with potential so early in exploration. But no way we stop here. Typically, I want to see 3 good but different takes on the same concept. I save this one as Option 1, and we go again!
“post apocalyptic village in fjord with circular gate, anime”
Step 1 — Due to how solid the first try went, I only add some words to describe what I think is missing, a “circular gate”. In our world, the purpose of Guardian’s Haven being located here is because there is a gate to The Lost Continent. (I pick V1 + V3.)
Step 2 — I picked both V1+V3, because they were both visually striking in different ways. V3’s results feel too full-on fantasy to me (pretty nice though). V1 has more mystery, I like the idea of the gate being elevated, inconveniently high up. V1 on V1.
Step 3 — Now, this is more like it. The rabbit hole went deeper on this one. It took about 7 iterations to find the prize. I’ll skip the picking details, but we eventually land on…
Step 4 — #1 of this batch. The composition of this variation gives me what I’m looking for. But the upscale is way too noisy. I try using the Remaster feature which lets the AI get more creative & loose with the stylization. We also try a Light upscale-redo which retries the same image but with less added details.
Step 5 — Remaster is actually mind-boggling to me. The same shapes, but now a skyscraping tower in the distance is a human-height garden gate. So cool, but not what I want. Light is better, but too abstract now.
Step 6 — I tell the AI to add back in a bit of detail, and we get the sweet spot. Details in the landscape+buildings, but no random noise in the sky. Save it, let’s try for 1 more.
post apocalyptic village inside of dragon skeleton, anime
I think we all have the idea by now, so I’ll skip the step-by-step on this one, and give you a greatest hits montage of notable variations that I went through to explore what this concept would look like if we emphasized a bit more of “what do the people do here?” instead of the overall setting of the world. People here hunt colossal monsters, stands to reason they might build their town out of those parts, right?
Here is the final upscaled image that we land on for this prompt exploration.
Finishing the Concept
Okay, so out of the three concepts, we ended up liking the second one. It had more elements of landscape and geography, while presenting a bit larger scope of civilization and more unique verticality of a magical element. We’re done right?
Not so much. This is a great starting point, especially given that I myself am not any kind of artist. I have a composition for the artwork, general structures and shapes for the thematic elements I want in the concept, and a color palette that I like as well. But I’m missing specific elements that tell the story I need to tell. I start off by documenting those elements relative to this starting image.
At this point, there are a few options — if I had all the time in the world, maybe I could photoshop my way to something halfway decent looking. I don’t have that, and I don’t want halfway-decent. I want the concept to look inspired and vibrant. So we hand off this image and notes to a proper artist and get their interpretation of this concept.
The proper artist in this case is Ohora, check out his Instagram (give him a follow too!). And turns out his interpretation is even more cohesive than what I had in mind, here’s a look at the sketched concept and notes on how he interpreted the piece.
After some minor back and forth, we land on a finished concept. What could have been a much longer and more expensive journey to reach this finished concept ended up taking a little over a week. The alternative to this workflow would have likely involved dozens of emails full of word salad and lots of human labor being discarded. Does that mean this is better? Not necessarily. But is it a great tool for our team at this stage? Absolutely.