AI map development?

[selden]

I have several questions and comments related to the use of AI to generate maps. My particular use would be to create a map of a generic tropical island. I've only just started my investigation of what's possible, so any hints and suggestions would be appreciated.

Questions:

1. To what extent is it allowed that members post AI generated maps on this forum?
I couldn't find anything specific in the rules. Did I overlook something?

Quite a few art-related forums emphatically forbid any posting of AI-assisted art while others have accepted it with open arms. (DeviantArt is one of the latter.) I'm assuming there would be "full disclosure" of such maps having been created using an AI.

2. What posts about AI map generation have been made?

I found only one brief post from about a year ago about getting poor results when using MidJourney. Maybe my search terms were insufficient.

3. Might it be appropriate to have a sub-forum specifically for the presentation of maps generated using AI?
One for detailed discussions of appropriate prompts, perhaps?

Comments:

Several free-to-use AI image generators have become available in the past year or so, although they tend to have limitations on subject material and the frequency at which images which can be created.

Microsoft's Bing Image Creator ( https://www.bing.com/create ) is based on DALL-E 3 and seems reasonably good at creating non-map images.

NightCafeStudio ( https://creator.nightcafe.studio/studio ) provides access to Stable Diffusion XL 1.0 and several older image generators.

So far my (small number of) attempts to create maps on these two sites have resulted in ludicrously pathetic results, but I suspect that appropriate prompts might be more successful.

[Redrobes]

Personally I see AI as a tool so long as it doesn't spit out cloned parts of other peoples maps. So the answer is more of a depends on the AI tool. But I think the genie cant be put back in the bottle and AI art is here to stay.

Also, I think maps, especially fantasy maps are probably the most difficult of types of art that AI can produce and to date I have not seen a single really good example of one. The AI tends to produce great art but terrible technical drawings and maps straddle the two. Also, fantasy especially, there is a very wide spread of art types for a fantasy map. From old school black and white line art to satellite imagery, some highly coloured, others drab or terrain type, then often with garish false colours, the AI has a wide band of source images to chose from which makes getting down to the core of what a good map is as a very difficult job. I.e. hard to train.

Right now especially, I would love to see some better than average AI generated fantasy maps that would pass muster in our challenges. All I have seen so far is just art. Pure art with no cartography in them. At this point I think maps would make a good captcha. I want to be proved wrong. I want to be impressed. Its bound to happen soon tho.

[Kellerica]

I personally am not a fan of the current trend of AI, so I'll refrain from commenting on it further in order to avoid coming across as super salty.

As for not being able to find conversations on these boards about it, I doubt there really has been much of it, which is why it's not showing up on your seach results. As Red said, maps are something that are (thankfully, as someone who makes part of his living drawing them) currently beyond what most AI can do. Hence the lack of discussion on the topic - as things stand now, it hasn't really had much to do with cartography and the Guild.

[ChickPea]

I was goofing around with a generator a few weeks back (can't remember which one) and I made a nice little map that I ended up saving because I thought I might try to remake it in my own style. It's a pretty little thing at first glance, though it falls apart the closer you look, but as a prompt for my own map I thought it might work (it's in the pile with five thousand other things that I'm going to make maps about...!)

I can understand the concerns of those who make maps for a living, but I feel AI won't replace an actual artist for those who have very specific requirements, at least not for a while yet. Though I can see how something like this would be more than adequate for people who just want a quick map for a D&D session. Possibly the makers of Inkarnate-type software are going to be more affected by the rise of AI maps than those who do commissions for exacting clients.

I don't know how I feel about people posting AI work here. It feels very low effort, not that there's anything wrong with that per se, but you could pretty much bang out a new map every quarter of an hour. I think people would get tired of seeing them though YMMV.

map.png

[ThomasR]

I almost posted an atrocious answer and had to exert a freaking huge lot of restraint not to hit send. So much for the salt

To be more productive, mapmaking is taylor made territory and image aggregators passing as AI these days are hopefully lightyears away from that ATM.

[selden]

My attempts so far have reinforced the comments above. For example, I cannot get Bing Image Creator to generate a traditional flat map. It continues to render the map's imaginary tropical islands in 3D no matter how many "flat" related qualifiers I use. *sigh*

[Wired]

To be more productive, mapmaking is taylor made territory and image aggregators passing as AI these days are hopefully lightyears away from that ATM.

This here sums it up nicely. AI generators at this point still are largely too dumb to grasp anything but the most basic concept (and even then, not always those), and I concur with Thomas that they're hopefully (likely?) lightyears away from getting any sort of details for maps right.

Also, the image quality and composition for AI maps really sucks once you look even a tiny bit closer at the generated results.

[Naima]

Everything I tried with the most popular AI tools ended up just in cloned and already seen stuff that usually needs a huge amount of rework , Midjourney , Stable diffusion etc, they now can create decent possible maps but when you look closer you get to see the several issues.

[Redrobes]

One must bear in mind that the AI were dealing with right now has no intelligence or any "concepts" or anything of the sort. It doesn't work towards a goal, cannot take in ideas or work towards a user specification. It is merely driven off of the prompt. The prompt text is a boiled down version of what other people have tagged images with or web pages with text near to that image. All the AI does is match shapes and colours to text based on whatever its been trained with. So it requires that all of the existing internet to have pictures which have a common feature before it will make a map with that common feature based off of the prompt.

So a penguin looks a certain way, an astronaut look a certain way, a house looks a certain way. But a map is a broad concept and fantasy is anything you want it to be. Thats why its currently having a hard time with it. Also, any prompt which is basically magic or abstract is going to throw it big time. Therefore AI in its current form cannot generate anything significantly new. It will always be a rehash of something existing and it has to be something that exists in multiple similar ways right now before it will lock on to it.

So use it as a tool to generate top down icons, or it will be absolutely great at generating character portraits or even heraldic insignia and so on because all of those things exists in abundance on the net. But it wont generate your map for you.

I think people give the current batch of AI much too much capability. Its really good at some tasks and seems magically good at some tasks that humans are not so good at but it doesn't understand the world. It doesn't know concepts and inter related thought processes because these are things which are not written down on web pages and it has not trained on them.

[ThomasR]

And it is ethically comparable to stealing a child's lunch