Art - Official Intelligence (an AI fever dream) by Ben Parisi
What do?
Find all 5 endings. Complete all 14 objectives. Do all the things to get 100%. But be aware you can't do all the things in one run.
Controls
Hold 'q' to see what you can interact with. Click on things to interact with them. Click and drag from your inventory to use your items on things.
What is this nonsense?
I used this Ludem Dare to assess the state of AI game development. That means I used AI as much as possible through the entire process. Design, programming, visual art, and audio, were all mostly done for me. In cases where AI generated content, I modified it as little as possible to fit the needs of my game. Basically cropping and masking in Photoshop.
I realize that, for many participants, Ludem Dare is about the exact opposite. They want to see what they themselves are capable of. I respect and appreciate this mentality. I'm very curious what people think of the ethics of AI, and what I have assembled here.
Ultimately, I was able to make a much better project that I would have without AI. But this was still a LOT of work. I was heads down for almost all of the 72 hours. And I'm truly proud of my efforts, and this game. I'm also excited to walk away from this with a framework for easily making point and click adventure games in Unity.
The ultimate honor would be if someone took it and made their own point and click adventure: https://github.com/BenParisi90/LudemDare54
What AI did for me:
- Visual assets were created in Midjourney.
- Photoshop generative fill to touch up assets.
- SFX were generated by: https://myedit.online/
- Voice over was generated by: https://ttsfree.com/
- Github Copilot wrote a lot of code.
- Chat GPT for some brainstorming.
What I did:
- Writing. (I wanted Chat GPT to do it for me. But it was doing awful. So I wrote all the in game text)
- Code architecture. Copilot writes good code, but it does not have a good sense of system design.
- Game Design. I laid out the world and game elements. Strung the quests and dialogue together with my in editor tools.
What worked well:
- Co pilot is amazing. The amount of brain cycles I DIDN'T spend on syntax was is impossible to understate.
- Midjourney makes great environments and reasonable paper dolls.
- Photoshop generative fill is pretty good at adding things to environments in the style of the image.
What was painful:
- Getting AI to consistently understand abstract concepts. Everyone faces the camera because Midjourney can't draw the same character from multiple angles.
- It also can't seem to draw a door that is open.
- And it has a terrible time making icons. Maybe 1/100 will be good.
- Image to 3d is a joke.
- Text to Speech is a weird landscape of shady sites that all seem to either cost too much, or be terrible quality.
AI Stack:
Github Copilot, Midjourney, Photoshop Generative Fill, Chat GPT, MyEdit,
Other Tech: Unity,
Post Jam Edit
Got lots of feedback that people assumed giving Neil the spoon was the only ending. - Added more obvious bedroom door. - Neil tells you to leave the room. - Door icon replaced with a more obvious arrow icon. - Screenshots on page of outside of room.
| Link | https://bigplayben.itch.io/art-official-intelligence |
| Original URL | https://ldjam.com/events/ludum-dare/54/ludem-daire |
Ratings
| Overall | 1058th | 2.871⭐ | 33🧑⚖️ |
| Fun | 1066th | 2.613⭐ | 33🧑⚖️ |
| Innovation | 906th | 2.903⭐ | 33🧑⚖️ |
| Theme | 1127th | 1.935⭐ | 33🧑⚖️ |
| Humor | 341th | 3.355⭐ | 33🧑⚖️ |
| Mood | 1011th | 2.7⭐ | 32🧑⚖️ |
| Given | 21🗳️ | 40🗨️ |
The astronaut Neil brought a psychic alien from the outer space and he turned out to be Mark Zuckerburg. Neil and the boy need to test the alien's psychic power by letting him bend the spoon, while analysing with the computer with the help of Sam Altman. But Whoopi Goldberg and Elon Musk smelt the danger of the aliens. Whoopi felt upset for knowing from the experience of startrek, human cannot survive from the alien, while Elon was expediting on the robot legs to prepare for the unevitable fight against the alien Zuckerburg.
A tool you might find interesting: ControlNet combined with Stable Diffusion can make people be in various poses, not just head on.
Good work!
Some things that would've been better regardless of AI: text speed and walking speed. Both were just a little slow. Did you try to get the AI to do any animations other than sliding the main character around? That would've been nice, although probably quite challenging.
I'm not a fan of the "dump a bunch of celebrities together and make weirdly boring fanfiction" angle but that sounds quite fitting for the kind of story you'd get with ChatGPT providing inspiration. I did nearly lose it at the hilarious line about asking Zuckerberg for legs though. Very on-point.
Overall quite a fascinating and different approach. The eye, boot, and door icons for activating things are all quite nice and generic usability stuff like that seems like a good use for AI tools in the future. I think I personally would hate using GitHub Copilot, which I've heard described as, "a kid who's eager to shout out the answer before it knows the whole question", but it's cool to hear the other side too.
If I would do this myself, I would probably pick a few models for different purposes. Midjourney alone is just too general purpose model and you will run into difficulties to get it to generate specific things on a decent quality (like icons).
But overall, I think the end result you have here is quite good! It still has that AI vibe but you managed to hide it surprisingly well.
@kinami Oh wow. This looks like exactly what I was looking for. Thanks for the tip.
@lovewerk Yeah, oops. I should have made it faster. Crossed my mind during the jam. But i just kinda... didn't do it.
@mduo13 Haha. Thanks for the review. I only hear bad things about copilot from people who haven't used it. My strong rec would be to at least give it a try.
@nash Yeah, someone recommended ControlNet. And I think that would have helped a lot.
@jamesw Thanks :)
The game itself could've used some QoL improvements, mainly text skipping and faster walking speed, music if possible. Still, made me chuckle a few times, congrats on submission!
But other than being a cool experiment it's not really that interesting, I mean it's obviously rough but it wasn't enough on the crazy side of what AI can give to hit the really funny bar.
What I'm wondering is why you didn't try out some music generation. Maybe you run out of time? It's still a lot of work.
As a stand-alone game, not so much. Ironically, the thing I liked the best was the fourth wall breaking dialogue, which as I understand, is the only thing not AI generated.
Regarding art-style, I find this to be quite fascinating. While each individual piece is fine, the style is very inconsistent and doesn't really "work" together. Not sure how to explain it, it's like you asked 10 different artist to work on this project without giving them any direction on feel, theme, setting etc. Not sure how to overcome this problem with something like Midjourney without some crazy prompt engineering. Would love to hear your take on this problem.