Im an artist working in games, and I absolutely agree. Lots of people think art in games needs to be "good" without knowing what that actually means. It's a lot more important that your art is coherent. Having coherent shapes and colors can do a lot. For example, just by choosing a color palette alone, you can create art that works pretty well.

Setting up any limitation will automatically create the coherence for a project. And you can go pretty minimalistic, too. Don't understand colors and light? Go black and white or sepia. Don't know about shapes? Use only one or two and design anything around it.

One problem with AI is that it doesn't use limitation as a tool and isn't able to contain detail. An indie developer who is inexperienced in art and able to manage their expectations doesn't have this problem. They can create naturally game art because they only know one way to approach it.

Or they use it to generate placeholder art "so they can get an an ideal on the final product while they're working on gameplay".

Super Mario 64's jumps were figured out with a cube bouncing around in an infinite plane. Their excuse is pure bs, good gameplay is good gameplay

Good idea. I'll tell my special interest magazine to make me more sympathetic.

You will not find a game engine without some AI tool.

I don't know where you getting this and spreading misinformation. I think Unreal Engine didn't have any Ai integration in its entire history. And I'm sure there are game engines without Ai tools integrated by default. I think the Open Source engine Godot in example does not have any of that. If I'm wrong, then please enlighten me. I mean seriously, I want to know if the engine includes Ai tools by default, because I care about.

People only care about AI when presented with it.

So they care about then? Whats really bad is, if companies or developers hide the usage of Ai and only admit using it after they got caught. There are many problems with Ai why people care about this subject. And it should be an informed decision of the buyer, if Ai is used or not or to what extend and what type of Ai. Generating art is not the same as autocompletion of words when programming in example. Using Ai to replace voice actors is also not something we want to see. Ai is trained unethically on data without permission.

If non of these games had AI generated visual elements, people would be non the wiser.

I don't understand this statement.

Godot is a game engine without any AI tools. At least none that I'm aware of.

Oh, right. The poison. The poison for Kuzco, the poison chosen especially to kill Kuzco, Kuzco's poison. That poison?

It's so weird how far self-censoring has gone. Now they're altering quotes because of this perception that it will be caught by the AI moderation...

Don't nitpick.

If NPCs can be dynamically fleshed out using LLMs, why not.

Sorry for the essay, but your "why not" got me thinking. I would argue it shouldn't be done, both for gameplay and safety reasons.

The application I see of this is something like city population in RPGs. Looks at Skyrim. Canonically, the cities of Skyrim were supposed to have populations in the thousands. But that wasn't possible to develop with realistic resources, and instead, they hand crafted a large, but still reasonable, number of NPCs to populate each town. It was enough to make the place feel like a functional city, but the cities themselves were physically small enough to make it all work. And, of course, like any RPG, after awhile you max out the dialogue tree of any NPC. This does cause you to lose the immersion.

So you might be tempted, "let's use generative AI to populate a truly vast metropolis. Let's build cities with thousands of NPCs."

You could try it, but it's already been tried. It's called Starfield. I have a weird relationship to that game. I find the plot vapid and empty. And there is no joy in exploration. There are innumerable planets, but each of them is filled with procedurally generated assets. Every planet is vast, fully and utterly empty at the same time. There's tons of bases, landmarks, flora and fauna to explore, but they're all repeats of the same thing, nothing like the vast yet still handcrafted worlds of Skyrim and Oblivion. There's some variety, but after playing for awhile, you see beyond the veil and the patterns become obvious. At that point, exploration loses all joy. I have a complicated relationship to Starfield mostly because despite hating much of it, I still have around 200 hours in it. Though that was mostly because I'm a sucker for factory games and got really into the base builder. The base builder, notably, doesn't rely on those procedurally assets for its core functioning. The parts I like best about Starfield were the handmade parts.

It's tempting to use LLMS to populate a vast RPG world. But soon enough, you will see behind the veil. Sure, they won't repeat the same catch phrases, but after awhile all the NPCs will start sounding the same. Instead of getting disillusioned because all the NPCs repeat the same 5 lines, you'll instead become disillusioned because they all sound like Claude or ChatGPT.

And worse, even if this doesn't happen, even if it never gets old, that's in some cases worse. Imagine you took this to the ultimate conclusion. Not only do you generate a mountain of dialog options for all your NPCs, you also embed an active LLM prompt window into the game. And let's magically assume that LLMs get good enough to never hallucinate and to always give unique and relevant answers.

Such a game might be legitimately dangerous to the mental health of anyone using it. People already get addicting to immersive games. Take a game as addictive as WoW at its prime. Now fill it with NPCs, each the most engaging conversation partner you've ever had in your life, each with infinite patience and willing to talk with you for as long as you want, at whatever you want, who will never question your ideas or find you at fault for anything. Each as unique as people in the real world are from each other.

That right there is a dangerous machine. That is not something anyone should build. Immersive games are already addictive to many. People are already falling in love with chatbots. Combine them together, and you're going to ruin a lot of innocent lives.

My guess is

1. It would be expensive.
2. It would be hard to control. LLM's are black boxes that often take you to unintended places.

It's a slippery slope. We have to object to ALL AI usage to ensure it doesn't get to that point.

I said metaphorically telling a story. Chess is literally a battlefield, but even physical sports focus on certain categories of human movement and form. They encourage certain body types. They require certain movements. Body language is a form of communication. The shape of your body tells stories of your ancestors and the story of what you've done and how you've treated it. There is poetry in body langue. It can tell a story. There are many kinds of stories. Not all of them are narrative. Some are metaphor. Every sport has its own vibe to it. Every game has its own feel. A video game is the creation of a human being(s). Another human being wants you to share an experience, whether game mechanic or literal plot narrative. Even a game with no plot at all still has heart, still has a soul. It represents another human being's expression of what they believe to be fun, enjoyable, and wondrous.

That is a deeply human form of communication. Even if it is entirely nonverbal. Every game played and loved represents the opening of one human heart unto another. And I find it morally reprehensible to be tricked into having that kind of experience with a machine.

I feel the same for most all creative arts.

My experience of AI image generation is that there are a lot of controls on style. I don't really see how art direction is more difficult, in the light of that.

I would be willing to bet that the people who were previously doing human made slop like you described are now the ones making the AI slop. They were already doing minimal effort; it makes sense they'd do even less when presented the option.

You're overestimating both how good AI is and how little effort a lot of these developers want to put in.

Generally, AI tools struggle with any type of specialized work and the same extends to game development. From my own testing, AI tends to butcher all but the simplest game coding tasks as a result of the larger and often very disjointed programming involved, and in terms of asset creation, it can create a lot of more generalized assets, but if you need a repeating texture or a texture for a 3d model, things immediately fall apart. That not to say it can't be done, but its a suprising amount of work for what is supposed to be an automation tool, and when compared to the old solution, is it really much better than just buying a premade game, and maybe swapping one or two things?

That said, my question is more that AI is ugly, soulless, and samey, but human art (and """art""") can be too. Do equivalent human works receive similarly lower reviews, or do the additional consequences of AI use actually factor in to people's perceptions, and if so, how much? For example, is a asset flip designed to rip people off going to be reviewed just as badly as a particularly soulless AI game, or AI going to be worse? Similarly, if you have a game with good programming and design, but AI assets, would it preform the same as something ugly in a human way like Cruelty Squad, or would it perform worse? Basically, how much of the negativity is because AI use has a bunch of negative effects, and how much of it is because the results of AI use are bad?

There are so many real issues with AI, but being angry about prototyping with it? That is a completely manufactured outrage by extremists who just hate the whole concept of AI rather than its misuse and mismanagement.

AI should be a public good that belongs to us all, managed, metered, and deployed in a way that is environmentally aware. A lot of good can be done, and has been done. We don't need AI girlfriends or surveillance, but we could really use it for research, data, prototyping, etc.. If you can't see this then ask some questions about legitimate use cases and I will share what I know.

Heck, it is objectively measured by a LLM adjacent seller like Faros AI.

The more LLM code in your company, the slower delivery and more bugs that you likely find on production.

Literally data is at 60% of daily tasks being LLM assisted, the throughput is (every value is an average) 500% slower, company delivers 10% less and the bug rate is +50% per PR and +250% production incidents.

At 40% of daily task being LLM assisted the bug rate was +9% vs pre-LLM.

Nuance is a word that doesn't exist in some people's dictionary, unfortunately.