AI In Game Dev

Category: Blog

I've been developing a game for the last 4 years or so. It's a city builder with colony sim resource management rather than money, and also is played on a spherical planet. 

When I started this project, LLMs were not very mature yet, but have exploded in their capabilities in the last year or two.

I haven't used any AI to develop any portion of the game, outside of using it as a learning tool to help me understand some concepts better. Even then, I haven't done this often. But recently, I feel myself gravitating more towards wanting to use it. 

Around September of this year, the engine I've been developing with Monogame, and the same one that will be used with this game, reached a point where everything finally started coming together. I was happy about this milestone, and thought, I can finally start making the game!

I have been doing that over the last couple of months, and it is coming along very well. I am still hoping for an early access release next year, and hopefully a Steam page set up showcasing the current progress by December.

The systems are coming along great. Your citizens will get hungry, they will die from starvation or old age, and they will go to work! Buildings are built on zones similar to other city builders but will be built only if you have enough resources, and only once your transportation vehicles have delivered the resources required to build them. Citizens are assigned as workers to buildings that collect materials, or have work trucks, and even material collection efficiency is directly related to the number of current workers / the max number of workers for that building.

I have done all of this without using any generative content in my game. I haven't used it to generate any code outside of some boilerplate. Every system, including my own custom ECS Duplecs (which I will open source shortly before or after early access launches), I developed myself through years of learning and practice. I know the speed at which I've done this could be vastly improved by using more AI in my development, but I see the vitriol associated with it, especially with artistic work like audio, images or 3d files.

Now, I don't have the same amount of knowledge with art or audio as I do with programming and development. It took me many years to get here, and it would be unrealistic of me to think I can make a decent early access version of this game with programmer art. So I have considered the use of generative AI for audio, textures, 3d files, and icons. Each of these things has already been mastered by generative AI, with the exception of 3d models. Before I dive too deeply into generative art, I wanted to see what it would cost to generate this type of art.

With images and audio, it's nearly nothing. I can do these on my home computer with a local generative model. Now with 3d models, I did experiment with local options, but these generate mess models that basically require throwing it away completely because they are far too many triangles. For this type of game, where you might have 20,000 buildings, they need to have very low poly counts. These local models just couldn't do it in a way that the model would be usable, and the artistic vision is lost, even when provided with detailed references or concept art.

This brought me to try out ChatGPT Astra, and the MCP for blender. This worked mostly okay, but still most of these models still ended up with more than 10k faces. These are still far too large to fit many comfortably on the GPU for processing, even with instancing. This isn't because of the file size, but specifically the calculations required for rendering. After playing around with it for a while, I came to the conclusion that I'm going to need to use my own hand crafted models. Even disregarding the concerns people have about the ethics of generative content, it's just not financially feasible and the models for this type of generative work don't really exist.

There does exist a middle ground. I found that having the MCP server running on blender, giving the AI access to that, and giving a colored texture showing what you want overall, you can ask it to produce a more detailed texture based on how you describe it, or with reference images. This is completely feasible right now, and would cost you about 2% of your 5 hour budget on ChatGPT using Astra. 

Results:

 

So now that I have found a workflow for 3d models, I found that I'm creating the model myself, unwrapping the uv's, and I'm creating a programmer art texture for them. Then I have the AI do some detailing to the texture to really make it stand out. So all in all, I'd say that yes it is feasible to use these tools in every step of the process to help with finishing this game.

But this experiment isn't an answer to the question, it's just an answer to if these tools can help me deliver my game in the timeframe that I want, without pushing it out into vaporware because the budget otherwise would make the game impossible for a single developer. 

The question is, is this ethical? I know I can do it, but should I do it? I'm still not sure how I feel about it. I know there is a huge negative sentiment towards AI in game development, especially with art and audio. But, even on Reddit, we can see that subreddits like r/aigamedev/ are taking off rapidly. So it seems there are people out there who appreciate AI being used to help develop games, but if you take a look on that reddit, I'd say 90% of these are "one-shot" tech demos from people who have no idea what's happening behind the scenes. 

Typically AI has been accepted into programming workflows, but these types of projects are generally perceived just as negative as generative art and audio are. The systems and behavior of the game will never truly match what the "developer" wants in these types of cases, because most of them don't know the types of systems that exist, to achieve what their end goal is. They lack the knowledge to guide the AI to do something explicit, so the AI does something implicit instead, and it might look like it's what you wanted, but in reality, it could be completely contradictory to your "vision". It can also prevent the game from being maintainable, because if you don't understand the system itself, you're not going to know how to fine-tune it, and you will just further prompt the AI to help, creating a cascading loss of your vision, just to try to mimic what you thought you wanted in the first place, only to have it be replaced by the AI's implied solution.

So given that the reception of generative content, in really all aspects of game development is perceived negatively, I'm not sure that it's something I should do. I would love to have great production level art and sounds in my game. I might be able to buy some sounds from existing packs, that would work with my game, but the art style I want something more explicit from. It needs to look how I want it to look, and this is much less generic than a click sound on a UI element. I'd love to be able to hire someone to make 40 buildings, 10 different trees, 20 different planet textures, rocks, and other foliage and decorations, but that is just not feasible with the $0.00 budget I have. I can't promise the game is going to be successful, so even a revenue split would seem unfair. 

This leaves me with a few options

1. I can make my own art (I rather enjoy the 3d modeling process), and have textures that are visibly programmer art. Once my game releases, if it's successful, I hire an artist and we work on replacing all of the programmer art and refining the look to more align with what I hope it can be.
2. I try to find some great artist who can do this on a revenue share split. This would be difficult, and I'd hate to make a promise to someone, have the game not make money, and then leave them as empty handed as myself.
3. I could use generative art in the development cycle. This could potentially lead to some backlash and players who otherwise would have played, maybe they don't.

In the end, after exploring some of these different paths, my mind is plagued with questions. As a programmer, I know how it feels to have AI disrupt your industry to a heavy degree. I know how those artists are feeling, and do I feel like my use of these tools in that process would encourage that same feeling in others? It can cost people jobs. We are seeing that now, especially in technical heavy positions, but who knows, maybe there is a correction on the way, and programmers and computer scientists are still needed more than we currently think. There is a possibility of public backlash, and that could affect the success of the game.

Do I skip using it just because of possible backlash? Is it ethical to think that if everyone were on-board that it would be okay, or is it still unethical because it's causing people to feel like they are obsolete in everything they are professionally because they can now be replaced by a robot, and not just physical labor, I'm talking about knowledge that took years for people to gain, or a lifetime to have an expertise in their area. The industrial revolution didn't take away jobs from people who spent their lives learning how to do what they do, it replaced jobs that took minimal knowledge and training.

I see this, and it does make me wonder, even if there were no backlash in these artistic spaces, that the usage of AI might still be unethical in some ways, as it completely devalues years of studying, education and practice in a subject without any recourse for those people. On the flip side, I will never be able to finish my game without either hiring people (with which I have no budget for), or making a game that is less than my vision. The worst case scenario is that the game never even gets finished. Would it be unethical then to use it since no one is losing out on anything anyway? I'm not sure the answer to all of these questions, and I don't know where we should draw the line. But, I am drawing the line on generated content for this game.

 

Created at: 2026-10-15 01:34:34.587959