Local Idiot Tries To Figure Out if AI Generated DnD Map is Useful for Actual Play
ฝัง
- เผยแพร่เมื่อ 16 พ.ค. 2024
- Sure, AI can be used for a lot of things. Having the imagination to come up with a fantastical realm to skulk through and plunder probably isn't super high on its list of strengths. That's my hot take, anyway.
Check out the map I reference here: www.wistedt.net/2024/05/12/fr...
You can also support the original artist here! ko-fi.com/pathspeculiar
Check me out elsewhere!
/ artofthecatt
/ artofthecatt
www.threads.net/@artofthecatt
artofthecatt.itch.io/ - บันเทิง
This kind of serves as the perfect metaphor that most art-related AI is only able to produce work at the very shallowest surface level. Art is more than just the act of physically creating it--it's the thought and decision behind it as well that make art what it is.
You can make a procedurally generated dungeon VASTLY better than you can make a AI generated dungeon.
The first thing that jumped out at me is that the passage is between the rooms are not clearly delineated. I'm still trying to figure out if those are narrow hallways or whether they're open to the areas around them and it's a raised walkway or what the hell's going on. This looks like it's done by an AI trying to draw hands. 🤣
Using modern AI to make an old school dungeon is hilarious. We could do that already in the 90s. 3e DMG has a random dungeon generator in it. I can still find a 90s random dungeon generator and sometimes I even use it. There's a reason they aren't the default, these folks are re-inventing the wheel here.
Maybe I am overly pessimistic but beside all the obvious ethical problems with AI I am pretty sure there will be a rather practical one: AI often reduces the time needed to "create" something meaning there will be an even greater flood of feces making it even harder to find interesting or useful things. Don´t get me wrong, I don´t want to gatekeep but this video is a perfect example that you don´t always need AI to make something accessible. One doesn´t need AI to create a map, all you need is a piece of paper, a pencil, some basic understanding of the game, some imagination and some time to playtest it. But that will become a problem: I am under the impression people won´t take the time to do the work and check if whatever the AI shat out is actually useful but just upload any crap leading to a situation - as one already can see - where the "I" in AI will stand for "incest". AI just "learning" from other AI produced garbage making matters even worse.
Very first thought: how tf do you get into any of the rooms on the AI crap? Robot forgot doors.
to some extent, "AI" generated maps are nothing new for d&d. I don't know exactly when it started but there have been map generators that just used plain old dice going back pretty far. a real straightforward method is throwing dice at a gritted battle map and drawing squares around where the dice land.
The twist is that high-tech AI is actually worse at making playable dungeon maps than what I just described.
This AI map gives details where they are not needed but doesnt give details where needed.
Honestly I feel Gen AI is reduntant here and there should be tools for procedurally generating complicated dungeons already.
(Non complciated are easy to generate using BSP from roguelike genre)
Preaching to the choir here but wow posting that wonky AI map was definitely not any kind of flex against the artist. If anything they proved the opposite of the point they were trying to make
I'm pretty basic about this stuff. I use AI maps, but I use AI maps to generate a foundation to fill in the blanks myself, adding or taking away things manually myself. It's like finding a shape model for a character, an animal, or even a mythical creature and adding in the eyes, the textures, or whether there's wings, or a tail, or any other number of things on that model. One day, AI maps could be amazing, and honestly I look forward to the day that the maps they generate can be used right at generation. However, even if the maps they design are perfect every time, there's going to be a pattern. Players are going to feel the presence of that pattern. Because of this, even if the AI generated maps become perfect, there will always be a need for that human touch to either perfect the base, or create the imperfections and personal feelings to perfection itself.
Respect to both sides, but even as an AI Map user, I weigh in for the human touch.
Is it really worse than "Tomb of Horrors"? Yes it is. Tomb of Horrors is playable as a "non canon playing session" where you laugh at what bad DMing and bad dungeon design is as much as possible by making a beeline for the not-fun-in-straight-DnDing parts.
Whoever pointed out the worse flaws of ToH earned a bowl of cashews. It was AWESOME.
Off topic but its great my algorithm from TikTok and TH-cam are finally mirrioring the same creators.
Oh heck yeah. I love this AI generated map. I am going to promptly use it as inspiration to make an actually functional map that has large sections of dark rooms and clear walls so that adventurers have a chance to plan out encounters in advance. This will let me give them heavily overchallenging encounters as they also have the ability to plan and prepare. Definitely gives a unique and eldritch feel.
I think I have finally found a good use for AI art: inspiring real art.
Dnd iz gud
The mapping system made by thousands of engineers working for years and with a multi-million dollar LLM did a better job than some dude throwing random stuff his brain vomited up. Sorry, but your bad take is bad.
ai is great at producting content
art ... not so much
if all you are after is benefit/cost then ai is unbeatable in most cases, getting a product thats 10% as good in 1% of the time is already 10x as much benefit per cost
so if you run out of content or your players run into a dungeon you don't have a map for, being able to just pull an entire dungeon map "out of your ass" and then do some corrections in ms paint for doors and some rulings for there being walls and so on? could be useful.
Then by the next session you can actually have a good dungeon map from one of the staircases in the original
I'll never not find it funny how twitter/x has such a grip on people who pretend to hate it, but act as if they enjoy it.
Now get the language model to add details. Describe each room and make it add the fun loot and enemies. Then you can even go back and forth asking "what if an adventurer does this/that?" And it will tell you everything. If something doesn't make sense, you just say "hey, it doesn't make any sense - fix it" and there is a chance it will actually fix the issue. AI now is a tool like any other, people get bad results because they can't use it properly. In 10-20 years we will get AI that does everything for us, but now it's more like "Google's I'm Feeling Lucky button that imagines things".
That sounds like plenty of unnecessary editorial work.
I can generate plot with few tarot cards (and a glass of wine) and it will have cohesive story without need to be proof-reader for bunch of blind monkeys with typewriters, who don't know how story structure works and what the plot was so far.
I've tried AI for in-game texts - the results were at best bland and uninspired. It was good for unimportant stuff (mostly fluff I clearly marked as not-worth-reading) but in the end I still spend my time editing whatever AI spilled out, so it would be somewhat usable and fitting with rest of the game. In the same time I could've write my own original stuff and be sure it fits the story and provokes emotions in players.
No time was saved and it was disrespectful to my players.
Why they should be bothered with reading something I wasn't bothered to actually write?
I once generated whole detective story for session with few 2d6 rolls and few randomly picked adjectives to describe people important to the crime. And it was fun, engaging activity for me and later for my players who could clearly follow clues and storyline I crafted for them. And I was sure that everything in the story was there for a reason I could explain in few words without missing a beat.
That's something AI can't do.
@@Fionor01 That's the whole point. Modern AI stuff is the hammer, not the blacksmith. People expecting AI to just create a full-fledged, engaging adventure from prompt like "dnd, fantasy, dungeon, fun", or create a good dnd map, are misunderstanding what the AI really is and how to use it efficiently.
It's entirely possible that a skilled DM won't need AI for anything. Just like a skilled programmer doesn't need AI to write the code. AI now is only helpful for things you aren't good at. You always wanted to make music, but you're tone-deaf? Try generating songs with Suno.
These things won't matter either way in about 10 years, given how fast the progress goes. If you asked me about this a year ago, I'd say there's no way AI will ever write a decent sounding piece of music. Now it's pretty decent. In future it will probably be more than decent. Same goes for dnd maps and adventures.
@@dzonydzas4964 It does the part of job I (and literally everyone else) enjoy and keeps humans do the part almost everyone hates.
It saves no time, provides no new ideas and makes everything more complicated to finish.
Technology that doesn't save time or work is useless. But it can save money, if you hire cheap unqualified idiot to fix the mistakes AI did - that's why CEOs love it. It's better for the bottom line.
Also, it won't get better. Those big models require shitload of specific data to plagiarize. There is not enough specific RPG descriptions and maps and over time the AI will be poisoned by trash - like "map" mentioned in the video, that will make whole model less usable. Adobe already has problem with AI being corrupted with stuff from another AI.
It's already pretty expensive to run AI model and this is just not use case that would make financial sense to develop and support. Even if it could do whatever you think it does. Which it doesnt.
@@Fionor01 We will see. I believe what we have now is more like a toy or a proof of concept, and it will get much better over time. There are limits to AI, as you said, but we are nowhere near reaching them right now.
@@dzonydzas4964 In that case... would you like to buy some NFTs? It will be really valuable in the future. It is on the blockchain. Pure futuristic stuff - it will even tie your shoes, walk you dog and pay your bills one day. Trust me, bro!
Breaking AI limits is VERY expensive even for general concepts. For special cases it is literally impossible. We call it AI, but it's actually really dumb model with actual limitations that can't be solved with enthusiasm and volunteer work. And Silicon Valley tech-bros know that. But that doesn't stop them in selling the crap out of it.