AI generated artwork is a hugely controversial topic for understandable reasons, and will remain that way for some time. But, like it or loathe it, it's here to stay.

It may almost be a cliché to say it at this point, but it really is a development as huge in its importance as the industrial revolution in terms of how creative works will be produced in the future, and artists are now faced with the question of how to meet the challenges posed by the existence of these universally available image generating tools. 

Join our Discord community and share your thoughts on this article with us!

However, we at More.is.More have learned first hand that as with any new tool or technological development, AI generated art is just that – simply another (if game-changing) spanner in our toolkit. In this blog-post we’re going to expand on how our team of developers – creatives, thinkers and techno-wizards – are working in tandem with their new artificially intelligent colleagues to create the artwork for Medieval.

As we all know, the speed with which the development of AI generated artwork has progressed is staggering.

In the course of last year we've gone from being happy to have managed to generate something at-least vaguely related to what we want, to being able to produce images that are indistinguishable from photographs. The latest advancements in terms of AI content generation have centered on delivering fine control over the results, and our current focus as creators is on just that –⁠ finding and establishing ways to fine tune the process to suit our needs, in order to create exactly what it is that we happen to be looking for. 

Stable Diffusion allows artists to train their own networks, offline, on their own artworks and art style. For example, this enables Illustrators to quickly iterate on concepts by speedily delivering a whole variety of possibilities for further development, or paths to choose from. 

Game Developers discussing AI Artwork

An artist can use a small number of his works to train a custom network called a LoRA, which will direct the diffusion process. This way AI Art generation becomes a tool for the artist, instead of his replacement. It’s important to think of AI this way, because no matter how good it may be at replicating already existing art and styles, there are many worries when it comes to AI. Its legal status is still up for debate, including how art for training the models is obtained or how copyright applies to AI generated images.

On the other hand, we’d like to point out how this same aspect allows smaller teams to achieve a much more ambitious goal without compromising on the artstyle. Scope is a key consideration when it comes to planning a deadline and team sizes. It’s easy to think of AI only as a tool replacing current jobs at existing companies, and we certainly don’t want to downplay this worry. However, as creatives, we also see the opportunity for smaller teams to create projects thought too time consuming, too expensive or simply just impossible to make due to other constraints. Human input will always be important, as without it, the resulting game may feel like it has no identity or personality of its own. AI is still just a tool, albeit a very efficient one.

A big point of interest for us is in discovering the possibilities AI art allows when combined with fine control.

To give an example, one of the things we wanted to do for Medieval was to create a number of illustrations in the style of illuminated manuscripts for the game icons.

We needed over a hundred icons, with each one having to communicate its very specific meaning – something that simply couldn’t be achieved with AI on its own. We discovered that using simple hand drawn line illustrations created in the appropriate visual style as a basis, in combination with AI to both color them in, and flesh them out with detail, delivered a fast and effective method, saving us countless hours of work while yielding fascinating results.

ControlNet input for “Game Developers discussing AI Artwork”

We achieved this using Stable Diffusion and ControlNet. ControlNet can use a wide range of inputs to direct the image generation, from edge-detection and line drawing, to depth-maps or OpenPose key points or image segmentation maps.

At first we needed to focus on finding the correct prompts to generate medieval manuscripts, as we need the AI to give us roughly the right result before we start trying to direct it.

We then fed the hand-drawn illustrations into Control-Net. Here are a few of the many icons we created, and the sketch inputs for each.

We think this is a great example of how AI generation tools can bring added value, and save time and money without compromising on artistic direction. It allows us to create more memorable art and exciting moments for players while artists don’t have to worry about compromising on their ideas or style. AI can be used to accommodate specific preferences and taught on specific art pieces, so its use will definitely expand in the future and affect trends in the industry. 

So there you have it – whatever the future holds, AI generated content is already part of the package, and we can’t wait to unpack all the possibilities it holds. As we’ve already pointed out, it’s not a key to infinite automatically produced content, but a tool in our arsenal. It’s important to learn how to work with it, just as it became important for accountants to use Excel and its functions a few decades back. Companies that rely on AI too much may end up paying the price, just like others have done in the past for relying on other trends or “surefire” ways of making bestsellers. But AI will truly be revolutionary and  avoiding it will do no good either. This short article is just a showcase of how we use it, but we’re excited to learn how you or other game designers experiment with this exciting tech!

Join our Discord community and share your thoughts on this article with us!