Design Philosophy concerning AI-Gen

2025 is almost over. The last thing I’ll do this year is finally write a comprehensive statement on my views on AI usage in my games. 

So… Like a year or so ago I added a big new set of “Danger events” Vidcom Illustrations and we also updated the Steam store page to better communicate that we are using AI generated graphics. This caused quite a discussion about LLM usage and a true little storm of criticism on some forum threads from a smaller but still sizable part of the SE players. 

I have refrained from going into direct forum or social media debates about the pros and cons of using AI generated art. I think anybody interested in such a discussion is already quite aware of the arguments on each side of the divide. 

Also I strongly feel that players have the privilege to vent their frustrations and viewpoints. As Voltaire said, and I paraphrase:  I will always defend your right to say what I don’t like to hear… I heartily concur with that. Furthermore I cannot really enter in a “thread-war” and stay civil at the same time. People call you a thief, and I say I hear you, they call me a criminal, etc… and at some point I would reply something insulting…  and I really don’t want to enter any mudfights. As we all know: the mud sticks on all the participants. 

However I have been reading all the negative postings over the last year…as well as the positive postings, sometimes even defending me. Thanks for that. It always brings a smile to my face. And also thanks for the criticisms, it did make me think. And that is always a good thing.  

I want to use this weblog posting to give a reply to most of what has been said and what the player base should expect from me concerning AI generation.

First of all… I love procedural design and Shadow Empire leans heavily in this direction. The emergence of LLMs interests me a lot as they are, I feel, in essence very close to procedural world creation. If I can use them to improve my procedural systems or increase the visual communication of my procedural systems I am likely to do so. The goal for me is to get the Planet and your little empire you are playing as immersive as possible.  

I thought everybody would agree with that… that for example for the event “alien critter infestation” an AI-gen illustration would be preferred above no illustration at all. Obviously I was wrong there. But I still believe this is better. Now, I just believe that not everybody agrees with me. I understand why some of you are allergic to AI-gen art and very strongly feel it’s an abomination and a soulless threat that threatens to devour us all. That it deserves a Butlerian Jihad! I understand, but I don’t agree. I still feel a middle ground in usage is possible.

There are also people who say that if I use AI-gen that I should make better use of it to actually have quality artwork and not people with six fingers, so to speak. This is something that I actually agree with and for the next graphics iteration, scheduled for Republica release, Q1-Q2, there should be improvements. And future iterations will probably be generated in the years to come when LLMs have made more big leaps. For better or worse it looks like the LLM generated stuff, their coherence among each other, and their eventual adherence to large reference portfolios, specification texts and contextual rulesets will only improve.

What excites me the most about the LLMs however is that it allows non-artists to create graphics and picture specific scenes they have in their minds’ eye. They have it in their minds’ eye, but they miss the artistic talent to visualize it for others to see. Well with AI-gen that problem is solved. And for this purpose I have done a major rewrite of the NewGfx Vidcom algorithms to allow unlimited modder-generated graphics to be added. This is not a simple case of allowing people to replace the existing art, which they could of course do, but the possibility of adding on top of it. 

For example it is possible for a modder to create extra Vidcom variants for all the scenes that are outside (on the Planet surface; not in a bunker) and get them rendered as jungles. In that case the new gfx files would need a Filevar attached that would permit them being used only if the planet has alien trees.  With these modder graphics added to the default graphics set the game would choose the new jungle ones if the condition set for it in the Filevar is true.

For example another modder could go so far as to make a specific jungle set with red lighting because of the Planet orbiting a red Sun. The Filevars system would allow it to only be shown if the alien tree condition and the red sun condition are both true.

Or for a more mundane, but extremely interesting, example a player could make a directory with say 100 paintings, and have it referenced from other Vidcom graphics then get random paintings displayed on a, say, interior wall or office wall. The painting curation and picks could be made conditional with the Filevar system.

I hope that some sort of collective graphics modding will become possible. It might provide fun moments to add to the immersion of Shadow Empire. I am not looking for unpaid artists here! As some critics might point out… But I am curious to see what modders will come up with. That is my main drive here. I like to be surprised. Heck, that’s why I like procedural creation in the first place.

For those that cannot stand AI art I am working hard on making it possible to disable all AI-generated art if the player wishes to do so. It is already possible for Portraits and Vidcom graphics and as it looks now this will be possible for the Stratagem art as well upon the next major core version upgrade and release of Republica DLC. 

Another thing I want to say is that yes… I could have hired human artists. Yes. But currently I am counting towards 200 Vidcom graphics and 500 Stratagem graphics upon Republica release. My budget just doesn’t allow that and neither does my time planning. As it would take any single artist a really long long time to draw all that. Not sure anybody would really want to do that job without an AI assistant actually.  The current approach also allows me to quickly generate new graphics whenever I do an impulsive addition of a new event or Stratagem.

So yes I thought about using human artists and yes I am sympathetic, but it would only work with a much lower graphics count. 

I am still looking for a way to invest part of my revenue, equal to my former “human” art budgets,  into real human artists and painters. We’ll get back on that topic after the Republica DLC release. In the meantime I can at least give the positive news that the 9 new music tracks that will be added with Republica are completely human made by our own Shadow Empire musician.

Now one more thing… I just have to point it out… the existing art that is being replaced was really not that super amazing. Most of it is actually copy-paste or collage-style-applied “stock art” and renders of stock-3d objects. I know some of you like its charm and I will try to continue to bring some new versions of this “programmer art” as some called it, to provide an alternative for AI-gen art. I will keep as much fallback option through the Prefs Tab as I can, without going to extremes though.  

I have been trying to please everybody, but will obviously not succeed 100% with that. So I am more moving towards trying to please as many as possible, including my own preferences, which weigh, not surprisingly, quite heavily, in those calculations of optimum pleasing. 

I think it is important to state my design philosophy concerning LLMs clearly.  I will make my games and their rules myself and I will never use the prompt “make me a game”. When (and if) that day comes I’ll move on to do something else anyway. Maybe take up teaching? take long walks? Plenty of options available!

Anywho, I am getting off topic. The bottom line is: The game and the rules and how instances of things (classes) are procedurally generated are all written with clear human intent by me, Vic. If I can however use an LLM to provide variety or immersion with, for example: an often-reused image, an existing class instance, an alien critter instance, a decision event of minor worker strikes or whatever sort of  thing like that, I will probably consider doing that. 

I aim to provide more immersion and if it is a choice between AI-generated/assisted immersion.. Or… no extra immersion at all… I will choose the former. 

That’s basically my core design philosophy concerning AI usage. Creative and mechanical direction rests with me. AI can be used to provide details and immersion, especially where otherwise none would be. I don’t know what the future will bring us….  and I do see how artist jobs are threatened by LLMs and I am trying over the coming years to offset similar modest sums as previous pre-AI SE graphics budgets to be allocated to human artist(s) so I stay more or less “carbon” neutral. 

I hope you all had a merry xmas and I wish you an inspired and healthy new year! Good luck to all of you! And I’ll be back in 2026!

Best wishes,

Vic