itch.io Community » Game Development » General Development

Is making an all-new type of game with AI okay?

A topic by iria-1342 created 13 days ago Views: 659 Replies: 22

Viewing posts 1 to 17

iria-134213 days ago(+1)

Hi, I'm a new developer working on making my first video game. I am planning on creating a very unique game spin-off an existing game genre (RTS) with mechanics that have never been implemented in any game before (I checked extensively). I originally wanted to build my unique game engine idea into an existing RTS strategy game code like OpenRA but it's too fundamentally different would be actually less work to build the entire game from scratch (per my discussions with Claude Code Fable).

Thus to actually bring my unique game concept to life, I am employing heavy use of Claude Code over many coding sessions to build the entire engine from scratch! Not even a powerful map editor like for SC2 would be able to implement my idea. However, because of the ambitious size of developing such a thing, I think the goal will be a minimal 2D RTS-type game with the unique mechanics.

So my question is: do people generally accept AI-coded games? or are they often ignored/disregarded as slop regardless of the merits?

I know there is a lot of backlash against generative AI and I agree protections must be in place to keep humans in the loop of game creation. If I ever make this game more fancy, I will seek assistance in the form of content creators for art assets, sounds, music, etc. But I'm doing allo the back-end-development with Claude Code. The end result won't be an "AI slop" game since I am directly managing the progress - but I worry that I will be labeled as such since I'll need to flag my game as "AI generated".

I won't go into the specifics of the game mechanics or why I consider it a new-type of RTS game (possibly sub-genre) now, but trust me when I say it's very unique and uses some advanced physics concepts. For reference, I have an extensive background in computational physics (my PhD is applied math) and my day job is a computational scientist at a lab. But my passion for video games is lifelong and I recently have started using Claude Code at work and have since transitioned to also buying a personal subscription for my own projects.

I'm thinking of hosting it once I have a working alpha version for feedback and ideas. I also want to look into how to protect my work (not planning on copywriting but I'll probably keep my github private).

Any other tips for new developers like myself who are using Claude Code for the entire programming effort, but plan on using humans for all assets (music, sound, art)?

Eldwood12 days ago (1 edit) (+2)

If it contains any AI generated content at all, including code, I refuse to try it and block the creator if they post in the forums.

DarkBloodbane12 days ago(+2)

You know, your post contains 2 different main points: AI usage and new RTS game concept. Unfortunately, you refuse to elaborate more on the latter which is a more useful discussion IMO. But fine, I'll play along.

do people generally accept AI-coded games? or are they often ignored/disregarded as slop regardless of the merits?

TBH, it's not easy to tell if a game is AI-coded or not. However, some ai-coded games have clear giveaways on being ai-coded. If people see those giveaways, get ready for backlashes.

People consider AI coded games and games using gen AI artworks as lazy work since the "devs" behind those games don't spend any effort in the creation process. Maybe those games do have merits but the AI stains are enough to keep people away.

Now allow me to ask you this question: you have lifelong passion for videogames, so why don't you use that passion to learn coding or at least learn to use an engine?

Irog12 days ago

Yes, making games is an excellent learning opportinity.

Game creation made me develop my programming skills !

KuroGamedev11 days ago

Funny. For me, it was the other way around. Learning how to build web sites helped me figure out programming skills for making games from the ground up, along with studying 80s computers and game consoles to get ideas on how to more efficiently organize and store my game data. Even though learning PHP was a bit of a "Really? You went with THAT?!" moment, it was still an important learning curve. Just... do not ask how my php game worked out, okay? Please don't.

It can’t be that bad. I remember those forum games like the pets you’d put in your signature. Those were the bomb.

BrotherCompost12 days ago(+1)

So AI told you you've got a unique idea that's never been done before, AI also told you it's so unique you'd need your own game engine. So you decided AI was going to code a new game engine from scratch, and you're going to oversee it without any real game development experience?

Bro just learn unity and stop talking to AI chatbots.

iria-134212 days ago

Unfortunately, I dont have the time to go full into game development. So I'm relying on Claude Code but with careful prompts. My goal is to make my idea come to life but if it's a good game in the end, I'll polish it with real assets made by real people (the backend is AI dev code). And yes, I truely believe my concept is novel but I don't want to spill the beans just yet.

I will eventually use Godot on the front end though. But thank you for your honest answers everyone.

BrotherCompost11 days ago

What I mean is it doesn't matter how careful you prompt, without a good understanding of coding and game design AI will always create a buggy mess, even when building within a solid foundation like unity or godot. It doesn't understand context, it loses its train of thought, it goes down rabbit holes to fix bugs and makes things worse.

Building a game engine is hard work, even for the most experienced game developers. If you let AI build all the back-end you're building on quicksand.

If you decide to go down the AI coded route (I still wouldn't recommend unless you're going to use AI as a learning tool and actually take interest in what it's producing), I would recommend you build some prototypes in unity. From what you've alluded to so far I can't see any reason why you can't build it in unity.

We're just trying to help, save you some time and money. Tokens are expensive, and what you're describing is going to waste a load of them for what will probably be a buggy mess.

redonihunter12 days ago

I originally wanted to build my unique game engine idea into an existing RTS strategy game code like OpenRA but it's too fundamentally different would be actually less work to build the entire game from scratch (per my discussions with Claude Code Fable)

The end result won't be an "AI slop" game since I am directly managing the progress

It might not be slop in the sense of a low effort project on your part.

So. The ai bot convinced you of something like this, did it:

There is this working gasoline engine powered car. You have this new idea of an electrical engine car. Instead of adapting the time tested and proven car design, where you know, it will drive on roads, have breaks, airbags and whatnot, it is better to design a new car from scratch. So you can implement things like retractable legs for offroad, a steel exterior, independently turnable wheels, airbags in the back of the seats, usb port on the roof, a mirror windshield since the placement of the airbags designates the safe driving orientation and so on. But hey, it will be a good concept, because the chatbot will even come up with anti rust fluid that gets sprayed on the steel parts.

I suggest you watch this video and dig up some numbers and evaluate how quickly you will reach the maximum of your claude subscription. Maybe doing it from scratch is not a wise plan. And maybe you might want to start with a less complex game than a rts. Maybe a tower defense game. Or a tower defense map in an existing rts engine.

iria-134212 days ago

I understand that AI can very commonly convince us of things that it doesn't find right away as "novel". I'm well aware of amateur mathematicians being convinced that they invented something amazing only to find it was known for hundreds of years in professional circles! I have indeed searched of similar concepts myself but I could not find anything close to the idea. And the reason for the entire engine rework for a "2D RTS" is that such engines are insufficient for my concept which "technically" is computed in 3D albeit using ring buffers and complex math optics. Basically every RTS engine out there functions in the same way, a unified map "state" that everything interacts with, what players see is what players interact with. It's all happening live on the same map (with some fog-of-war to hide stuff). My twist comes from replacing the fog of war with something far more complicated.

I'll probably share the concept in another topic since this topic was more about the AI sentiment regardingg AI coding (not related to AI art/music/sound assets). As for whether my game concept is "fun" is another question as well, but I am sure it will be unique.

iria-134212 days ago

And thanks for the video and suggestions. As for the "why RTS" and not say a tower defense, well it won't be as interesting with my concept. My concept revolves around motion, strategy, and positioning to get to appreciate the special effects to their fullest.

redonihunter11 days ago (1 edit)

I did not mean you should use your unique mechanic and implement it in a tower defense game.

I suggested you should start creating a smaller game first. Right now, you seem to have zero game development experience. It is that saying that one should learn to walk first, before running. There are jams that have ai games as a topic. Maybe try joining one of those. This is Itch after all.

A tower defense game came to mind, because it shares a few things with rts, on the mechanical side. In one of the simplest versions, you have something like Plants vs. Zombies. But you can go arbitrarily close to a real time strategy base builder game, where you have workers building towers and could spawn your own army to capture the enemy base.

My concept revolves around motion, strategy, and positioning to get to appreciate the special effects to their fullest.

Uhm. You sure you are not reinventing a complicated type of tower defense game, but for multiplayer? There are loads and loads of different mechanics seen in tower defense games.

https://duckduckgo.com/?q=Desktop+Tower+Defense&ia=web https://duckduckgo.com/?q=whiteboard+tower+defense&ia=web These had you build the paths by placing your towers.

Anyway, you sould split the ~~problem~~ project into smaller steps. And one of those steps is to verify that you can actually build a game with the intended method.

And if you are confident there, you should tackle again the issue of new engine versus modification of an existing engine. The problem with rts game is, that they are multiplayer. And multiplayer is a nightmare for development. Big companies often can't get it right and I have seen it happen that they scratched multiplayer features for release, because they just could not get it to work reliably.

Bringing a new mechanic into an existing framework sounds orders of magnitude less work than building such a framework from sratch. That advice that the ai bot gave is highly questionable. Building a rts is years of work for multiple people, but the ai is confident it could do it in months with supervision by an amateur developer...

masterofgiraffe11 days ago

Your game engine is going to be unusable and buggy as hell if you're relying entirely on AI for it. And your game is not going to be unique if it's being created by a chatbot. AI is not capable of creativity.

lol

Evolutionary Games11 days ago(+1)

Search engines don't play our games - therefore, they don't know much about the game itself and cannot answer a question like is this specific mechanic unique. But take care to research your target market - people who pay point and click games might not want to solve an intricate puzzle to continue playing a point and click game.

If AI doesn't do a specific feature right - or can't do something you want - you're going to be stuck. You could spend a lot of time and money and end up with nothing. Whereas using game-development software gives you access to forums and experts that could help you out.

I've played a few AI created games that were fun. I've played more that stopped working after a few levels or were unplayable because the person didn't think of the little things that make a game playable - for example, you'd fall off the edge, it was too fast, or it became boring and repetitive very quickly. I think in these cases the developers were relying solely on AI, and AI let them down.

So, proceed with caution. Make the game if you're going to have fun creating the game. Don't make the game if you expect to make a good (or any) return on your investment - the odds are stacked against you.

redonihunter11 days ago

cannot answer a question like is this specific mechanic unique

Yes!

They are trained on material where people would talk about games. And players rather seldom talk in detail about specific technical game mechanics and how they are different to other games.

And they might not be able to recognise a concept by another name. Though sometimes they are very good at splitting concepts into their essences. It's after all how they work internally with all those tokens. They are made for processing language after all.

iria-134211 days ago

Okay, thanks for the additional information. I did extensively research various RTS games and the closest to mine in terms of strange mechanics would be maybe Achron or Nebulous Fleet Command. Again, I realize that building a game from an existing back-end engine is the ideal goal. But my project requires a special form of game state mechanics that requires storing space-time state vectors for every single vertex in the game! All the common RTS back-ends are fundamentally incompatible because of how the game state is represented.

I'll make a separate post soon to discuss the concept and where I am so far, but I was the one that came up with the concept and implementation strategy, not the AI. The AI is handling the boiler-plate coding and will help with the Godot integration. The art assets are minimal for now (no generative AI art/sound/music to be used) - just basic shapes and colors.

iria-134211 days ago

Also, I should add the game is 100% not for profit. I am doing this with my personal funds as a hobby.

DREAM_SEARCH_REPEAT10 days ago

Many people will just hate your game without even trying it because they are too blind to understand that AI is just a tool. But there are others that don't give a cra* about how the game is made as long as it's good. You are not less developer or artist or anything by using AI.

Being said that, i'd like to point that your project sounds too ambitious to be a first project. I'd like to recommend that first you make small games first and learn about the things you wanna deploy. I mean, learn programming, game design, etc. before asking the AI to made the code for you. That way you will have the technical language to ask things properly and the knowledge to find errors the system will make (and trust me, it will make a lot.) and fix them yourself. Also, the knowledge will give you more control and authority in your project.

Basically AI is to speed up things you already know how to do, not replacement of tools/professional/knowledge.

Basically, responding the title's question. Yes, it is okay.

iria-134210 days ago

Okay thanks for the warnings! I'll keep that in mind. Yes, I'm using AI but only for bringing my creative ideas to life as a tool. I agree that it's ambitious but it's something I don't plan to release anytime soon since I totally agree it's a lot of work!

As for progamming, I am a computational scientist and have extensive experience with Python, C++, MATLAB among other languages. I am well versed in computer science terms and game theory concepts. I have deployed other projects (non-game-related) via GitHub for work so I'm not new to software development. For the parts I don't know, I use my gaming knowledge and ask people; however, being able to google algorithms for things helps a lot (e.g. flow fields vs A* for pathfinding efficiency).

AndruxaSkroz7 days ago

Hey Iria, don't get discouraged or threatened by anti ai peoples :) did you try pygame with python, i use it a lot for prototyping and db provisioning, unit testing? C++ is always good for back-end stuff; and if you are a computational scientist with good practical application of statistics you can use that knowledge for optimization, of npc behavioral models, and many other things etc. I have very extensive AI workflows i think codex is a must, comfy for local models, ltx 2.3 workflows for movies, ace step 1.5 for music. When it comes down to ai coding in my opinion AI is still amiss, its to unreliable as a dependency, and it is also unreliable on large backend architecture with thousands of files, API calls etc. For assets i still spend hours, days trimming , cropping, redesigning in krita and blender. 3d AI models require retopology for most commercial ones which can burn days, even with blender mcp or blender codex still way to time consuming, unless i am doing something wrong lol. For AI movies and music need decent loras from hugging face and decent workflows and yet still need good seeds to generate content. Also AI is pretty poor for security and for niche tools that dont have a lot of available data in public access. So there you go , maybe some of this will be helpful )

iria-13427 days ago(+1)

Thanks for the tips. I am using ClaudeCode exclusively for the coding in a neat way. I'm the architect with all the design choices, Fable is the foreman/civil engineer, and Opus/Sonnet are the contruction workers. It has been a good workflow so far and everything seems to work since after every few milestones, I do a play-test and report my findings (bugs, suggestions, ideas). Then I make a new batch of milestones based on the feedback and continue onwards. I do use my computational background to suggest certain algorithms or ways to implment ideas but often times Fable already knows of the method first and proposes it! I haven't encountered anything as of yet that wasn't solved with Fable and the other agents (maybe my prompting was careful enough). And I know the code works correctly because of my direct playtesting (I know exactly what certain effects should look like due to my physics background).

Since I'm committing to not using generative AI assets beyond simple shapes/textures/etc. the UI is minimalistic but functional and there is no sound implemented yet. While it is entirely written by AI (in typescript for playing in a browser), it runs quite well and can be adapted as needed (e.g. I added debug menus, unit parameter editors, runtime metrics, etc.).

My game is strictly 2D-only due to the computational complexity behind the game mechanic (which I'll share soon in a different post) and while C++ is an option for faster execution, this isn't needed anytime soon. But thanks for sharing your experience with AI tools.

Log in to reply Join the discussion. Log in or create a free itch.io account to continue.