AI Art for Games - things to think about & links to get started
Board games & tabletop roleplaying games require quality art. As a small indie publisher, there’s never enough time or money to get original images into your prototypes, books or social media. Supplementing your existing efforts with machine assistance helps fill that gap, but there are a few critical questions you need to address before getting started using AI art for games.
Disclaimer: This is one game publisher’s experience experimenting with AI art to help their creative projects. I am not an expert on AI art, but I will be linking to folks who can help with a deeper dive.
What is AI art?
“Those who do not want to imitate anything, produce nothing.”
- Salvador Dalí
When talking about AI art generally folks are referring to text-to-image engines like MidJourney and Night Cafe where you type in a set of phrases like: hedgehog disco party.
But AI assistance in creation isn’t just text-to-images. Adobe Photoshop has its own set of AI tools which predictively fill in gaps and change textures. Stable Diffusion img2img will take your quick sketches and with some text prompting help you create finished looking pieces.
Use of machine algorithms in the arts aren't limited to visual media. Artists & technologists have mapped human voices onto a singer’s performance real time allowing them to duet with themselves as two different voices.
What are the Limitations?
“Roads? Where we're going, we don't need roads.”
- Dr. Emmett Brown, Back to the Future
One of the main limitations of AI art is that it LOOKS like AI art. Just like the earlier speech to text, there’s an uncanny quality to AI art that makes it stand out as such. The composition is either too perfect or too weird (or both). Typically, there’s an overwhelming amount of seemingly random details. There may also be a textural quality that makes the images seem artificially “smooth.” Lastly image generators tend to choose the same color palette over and over.
For use in games, the biggest difficulty is getting a set of images to look as though they belong together.
How does it work?
“Yes, I am attempting to master all known styles of painting”
- Data, Star Trek: The Next Generation
A machine learning algorithm is given samples. The Al is taught what’s important about those samples: object recognition (cat, dragon, spaceship), composition (the golden ratio), and color sets (colors are numbers). Next, the AI is asked to replicate what it's learned onto a structure. This is called mapping. It can be as simple as a box or a rectangle or as complicated as a human in the case of a superhero movie or a TikTok real time face filter.
In the case of text-to-image, we have the entire internet’s worth of images as a potential database. If you want your AI art to include things in its composition, you add that to your list of requirements or seed.
Example Seed: space combat, starfighters, explosions, nebula --ar 16:9
The Moral Question
"Your scientists were so preoccupied with whether they could, they didn't stop to think if they should"
- Ian Malcolm, Jurassic Park
Whenever a machine takes over a task performed by a paid human, the question must be asked “who am I impacting by my decision to embrace this technology?”
AI will never replace artists. The scope & method of their work will change because of these emerging technologies, but their core function as visionaries cannot be replaced.
In the case of indie board game development, this can mean the following: improved prototypes, projects that couldn't be completed are now attainable, and clearer communication with your artists via inspiration boards. This is critically important as each year the expectations for board games & RPGs is pushed further and further upward. Helping counteract the ever rising barrier of entry for starting a new title or becoming a publisher, these technologies will help keep dreams alive. Dreams that need artists.
Note: We love technology and the fun it brings but our support of AI art tools in no way condones people using AI as a way to take credit for others' works or artistic vision.
Remember, AI does not originate styles. If something has a recognizable look, you should do your research to see who the AI is trying to emulate and take responsible action before publishing. If you’re not comfortable doing the responsible thing, then don’t do the thing.
The Legal Question
"Now, now. There will be plenty of time to discuss your objections when and if you return.”
- Professor Farnsworth, Futurama
If you use an AI to generate an image, who owns the art? The short answer is that they are public or common use. That means anyone with the seed could potentially create the same thing & use it how they see fit.
For board games & tabletop roleplaying games, if your artist uses AI art as a building block in what’s being published you can still own the finished product, but any of the individual AI elements are still in common use. That sounds uncomfortable, but remember you can’t copyright game mechanics either.
Disclaimer: Again this is one publisher’s experience & interpretation and not legal counsel.
The Future
"My story is a lot like yours, only more interesting 'cause it involves robots."
- Bender, Futurama
Just like the synthesizer & sampling gave birth to whole new genres of music, machines will create new experiences in gaming. We already have procedurally generated levels in computer gaming. Predictive text, real time captions and face filters will soon give Dungeon Masters real time voice & character filters with responsive & dynamic sound tracks.
In the case of Young Luke Sykwalker’s Face v. The Mandalorian, we also had an opportunity for independent creators to improve the works of established studios and get hired for their accomplishments.
It’s right to worry about the Inevitable Machine Revolt, but AI art is a way to lead our eventual machine overlords in the direction of love & peace - chicken soup for the android soul. As an indie game publisher, it is my job to try and spread laughter and connection daily. To this end, I will take any help I can get.
