AI's impact on your studio might not be as beneficial as you think.

Part one: Efficiency

January 17. 2025

Open Graph Image Illustration

This essay is part of a multi part series about the impact of LLMs on game production with a strong focus on creative and art production in the games industry. But I'm sure most of the reasoning will apply to other environments as well.

AI systems, or rather LLMs are taking the world by storm. Regardless if society wants it or not, these systems are propagating into our lives, are put into products and software and are used for all sorts of tasks. Having worked in the entertainment industry, since 2007 I have never witnessed such profound and rapid change occurring. Right now, for many, this change is life altering in a rather negative way. LLMs as they are "applied" to the world right now are bearing many problematic traits, not the least of which is the environmental impact training and usage has as well as the cost of resources and of course the legal discussion of whether training on copyrighted material should be allowed. Also AI provokes deep changes in our society at a time where poor and rich drift apart at an ever accelerating rate.

These points are valid, and personally I think, more than enough to stop the race. But I don't want to write about this today. I want to talk about that one other promise AI-companies are quick to give to their future customers: More efficient, better quality with less human resources involved. To some, that might sound good. But is it?

Leading up to now

In the last few years the technology around ML Systems like LLMs accelerated at a rapid pace. A growth that can be attributed to astonishing technological advancements in the field of real time processing powered by large clusters of GPUs. The defining traits of LLMs is that they are uniquely suited to process large volumes of data and find similarities based on certain preconditions.

Doesn't sound like much to the average mind, but once people realised that pixels of any image are just data, things changed. An image started to emerge. Many of them indeed. And with more training, more capital and promises of profit, training on ever growing datasets became possible. Stunning technology, but no Artificial Intelligence. No reasoning, still broken hands, strange details. Good enough in some cases? For many companies it may seem that.

So what are you reading here?

I will disclose that I am strongly opposed to how things are handled. But, as I said above, let's leave that out of the equation for a moment. I am actually interested in what this technology will do to game- and especially art-production in entertainment. And there is a lot to unpack. Also, being opposed to AI does not mean I haven't spent some time evaluating the technology. So from the early iterations of Midjourney to current gen systems I have tried most of them. Not productively, but for research. No AI output has found its way into my work. But believe me when I say that I did try to make something out of it. Just to see if I could. So let's continue and explore the first promise.

The promise of efficiency

Efficient work. It has a certain ring to it, threatening or promising depending on whom you ask. In studios around the globe it usually means: "get stuff done fast". Getting things done with as little friction as possible.

It is a nice term that can be used for well defined tasks. If my task is putting an orange from the left bowl into the right one it can be measured really well. Maybe 1.5sec on average. If a task is less well defined, this measurement tends to fall apart. Design me a character. We know little about her so we need to start concepting and exploration. How long will this take? Maybe the first sketch nails it. Maybe the 200th does? Maybe it gets rejected by the director. Maybe the story changes. Maybe, maybe. You simply don't know.

You can have a gut feeling based on experience and on how well you know the team and the environment. And here you have two things that scare people who have never worked creatively like hell: You have to trust people to know what they are doing AND you don't really know or can understand WHAT they are doing. Creative work is... difficult to manage in that regard. You have to improvise a lot. Especially in planning and execution. So what every studio exec is trying is to put things into numbers and measurements as much as possible. Since those measurements are seldomly right, it puts pressure on everyone. Pressure in turn burns people out if applied too much and too long. That's why late projects tend to lose/switch people and why mismanaged projects will only rarely recover and mostly crash into a burning wreck.

Enter LLMs, The saviour. You are promised lower monthly costs, a much higher output, and a way better measurement on how long things take. Less volatility while saving money. True? Burnout is a serious personal issue of course that you should take great care to avoid happening in your team but it also has a profound financial impact on your project:

First of all, if people burn out on your project it is THE most expensive thing that can happen mid production. Not only will your output go down considerably. You'll lose valuable knowledge, you'll have to find a replacement. This will cost your leads time. And time is money. After hiring the whole onboarding has to commence. Which will again cost time. The person needs to ease into the team and your processes which takes time away from them. And then there's still the prospect that you might be losing someone new after a short while because they aren't a good fit.

The more responsibilities you put on people within a short time bracket, the more pressure builds up, the more likely you are to burn them out. But this can't happen to a prompt engineer, or can it? Since you put so much emphasis on efficiency and a high volume output you are actually bundling a lot of responsibility and dependency on a way smaller pool of people than before. Think of it this way: If you replace 9 artists by three prompt engineers (prompt artists, whatever you want to call the job). If one leaves or burns out, you are losing a third of your team. Your ratio is building a very scary situation there.

Maybe you are thinking "But! Of course "AI-Artists" don't burn out the same since they have less to worry about. 200 new images in two hours is no problem right?". Let's dive a bit deeper into that.

Creativity and the human mind

As humans we are able to or process audio visual information efficiently only for a certain time at a stretch. This time is drastically reduced when we experience Information Overload which is extremely common whenever we watch rapidly changing visual information. Like games, or movies. Or in game production when we browse through our library of reference images. After some time we need rest to recover and be ready for new information influx. Sleep helps as the brain processes these visual influxes in a sort of queue. Under stress it takes days for the brain to catch up.

If your job is to think about, generate AND browse hundreds of output images an hour you'll arrive at your exhaustion level MUCH faster. It's just natural. You wont stop to "function" immediately. But you'll lose mental attachment and your ability to make informed/creative decisions. The pressure of efficiency is directly opposed to the essence of creative work. Creative work needs time to be mentally processed and these processes cannot be accelerated too much. In that regard it does not matter if we create art with pencil, camera, photoshop or LLMs. If anything, working with an LLM will wear us out even faster.

Another effect we have to take in consideration is creative fatigue. A term which is used by ad providers to describe the phenomenon that we tend to get bored really quickly if we see the same or similar things over and over again. We get bored by the hight volume of images that look so samey everywhere. The same warrior girl in our art feeds, the same art styles, the same characters. We get bored of the same music and we often find it hard to read part 4 of any series unless it is captivating with big changes and ever growing pool of mystery. Part X of our favourite game often fails to provoke that great sensation we had on part one or two.

This doesn't only pertain to your audience it is especially true for your prompt engineer who will need to generate and select from thousands of images, with the same style over a very short amount of time. There will be next to no attention from a certain part onwards in the process. For me this was reliably at the 30(ish) minute mark when trying to work with LLMs. Your mileage may vary. But the quality of your output will be worse or just random junk after a certain time span.

"But! The amount of images we get will practically guarantee a hit on far less time and manpower! Right?"

The output and the cost

Unempathically disregarding your now tired team member for a moment you might want to look at the output. To the untrained eye it might look enough. But then you have to take whatever it is you have and kinda fit it into your project's world and visual identity.

Snug fit? For the most part I can guarantee it will be a force fit with a lot of work involved to clean up the damage. I understand that each project and each vision keeper has different standards. For some projects "something" is better than nothing. But you'll always be lacking compared to projects that have a strong, coherent vision and a story to tell in even the smallest parts.

Why would that be? We get images, music, stories, everything from the LLM, don't we? Well, you get many parts that won't fit well together out of the box. Not even remotely. I will discuss the impact on art direction later in this series. But for now, in order to create something out of those parts you'll have to put in a lot of work. Both in writing and in art. We need coherence for believability.

As human beings we don't have to be experts in everything to get certain things subconsciously. The famous Uncanny Valley Effect is just one of the shapes of this. We kinda know how people, houses, ships, breadcases and pistols etc. should look like subconsciously. All the beloved stories in media have been told by people that were positively driven in terms of their attention to detail. It takes the right mindset and a lot of time to get this right. While it might not be necessary for every project to be this diligent, visual consistency and a recognizable artstyle go a long way.

In no world you'll be able to get all of your production right from LLM output. You can make something out of it with a lot of artistic input and work. But for the most part the raw output is just not there in terms of quality. And I haven't seen a demonstration that won me over. Especially not when it comes to 3D. But even for 2D it is already highly rare that an output would be good enough even for a basic amount of polish. Just look at he ire fans are giving unchanged an badly monitored AI output Six finger Santa at IGN.

Monitoring means there still needs to be high level supervision (which is expensive) AND a competent artist on board that would be able (and willing) to fix sloppy AI output. Are you actually saving money then? Especially for small teams it seems unlikely. Lead artists or directors cost more money and their time is more valuable. If they are occupied with monitoring and fixing outputs you are bleeding money. The issue is made even worse if your directors manage with the help of AI.

"But! We can finally fix this with an artist on board, can we?"

Speaking of artists

Getting artists to fix these outputs is a very tough topic. But generally spoken you are breaking the basic premises of good human resource management. Let me elaborate.

Artists are driven by a life long pursuit of learning, improving and following a path that requires a lot of passion. They can probably fix the six fingers because they know how to paint them in the first place. But they will get bored really quickly because their work gets reduced to assembly-line work. For motivated indiviuals this practically guarantees mental fatigue and facilitate burnout much faster. Burning out people is costly and also it will definitely give your studio a bad reputation. The world of entertainment artists is small an well networked. They won't also get portfolio pieces this way.

Another issue you should be aware of is this: The fundamentals of Art, Design and human vision haven't changed. They didn't change just because LLMs are suddenly there. They won't change for a few hundred thousand years at minimum. Neither does our perception of other humans, art and stories. Neither did our mental states change. So you'll still need to manage a team of humans that should know how good art can actually get through to your audience and captivate them.

LLMs consistently fail to do that. They'll probably fail at that forever. Sure, at first glance that new AI children's book looks nice. But the cracks show very quickly. It doesn't feel right. It doesn't take us on a journey. It doesn't look that great. Same with books and art. There is one thing missing, and that is intent.

In short that means that you still need artists and writers that went through the gauntlets of classic education, self improvement, networking, learning and endless fascination with the world. You still need engineers that know about human machine interaction and information theory. By letting go of a chunk of that knowledge go, by laying off people from your team, you'll have a hard and costly time to recover that information.

Speaking of recovering that information. Right now, prompt engineers know nothing to little about this and even if they do, they cannot change the output reliably to meet your needs. Realistically, if you start teaching that knowledge to newly inspired prompters, it will be 10 years plus before any of that will be remotely useful to you and your projects.

The verdict

All of the above directly infer costs to your operations. Costs you were promised you could save by embracing the new technology. While there are uses, in a creative environment I cannot see a good use and a good cost/risk relation involved. Your costs will just rise in places you might not have expected with a lot of expenses going to LLM service providers. And those costs are going to rise, especially with capable talent missing from the pool in the near future.

So Look carefully at your AI endeavors:

I'll write about other implications in Art Direction, Art Education and Project Vision in later parts of this series. See you there and get in touch on LinkedIn if you want to continue the conversation.

written by Richard Schmidbauer Richard has been working as Lead Artist and Art Director for indie games since his debut in the industry in 2007. He has been a gamer and a creative person his whole life. But part of him is driven to science and tech. That's why he sometimes tends to think about things in a rather analytical way...