Use of AI
What is your stand about it ?
Personally I find it can be a valid support, can help enhance creativity and speed up processes , but requires to be reworked and built upon rather than using it straight away. Is this the end of artistry or a new beginning?
In the end is this the future ? Should we adapt or fade into oblivion?
There's clearly a lot to say about this and we have had a few chats about it already. The last link on this list has a few fantasy maps made with it. I think they are sort of clones of other maps and have some issues with them but it was an excellent effort. Whats going to be interesting is how much better they get as time goes by. As a software person my job is on the line too and whilst I have seen a lot of talk about how well it can program simple things I have yet to hear how its done a great job at something complicated. On the other hand, I have heard an inexhaustible supply of tales of how bad it did. So right now its an aid to many tasks but wont replace them just yet.
https://www.cartographersguild.com/s...ad.php?t=51132
https://www.cartographersguild.com/s...ad.php?t=51532
https://www.cartographersguild.com/s...ad.php?t=51950
Keep posting sample AI maps tho in these threads or this one. Lets just see how good they are and how much time we all have left !
I am not considering replacing map making per se but eventually to use for making complementary elements in graphics eventually that could be sourced from.Originally Posted by Redrobes
I'm right there with Naima. I don't use it to make a map (for one thing, it can't. For another, then I wouldn't have any fun.) But I do use it to create border elements, which I tend to be bad at/not interested in, and making interesting city icons, compass roses, etc. I hardly ever get something useable right out of the box, so normally spend some time, sometimes quite a bit of time, making these elements workable on a map.
Most of the current synthesis models are good at elaboration (it's what they do: searches from the current pixel context in their encoded database of prior art to find something similar). The basic operation of most of the image generators is to start with a low-res noise image and some areas of the database to explore (the prompt) and then explore the database in the specified area to elaborate on the current image. Repeat until converged.
If you start with "make me a map", you'll find that there aren't really a huge number of basic maps, so they tend to look similar (even more similar than when humans do it). If you start with a sketch, then it can elaborate on that sketch to get something new based on the details in somebody else's work. The question has always been regarding how much human input does there need to be to get something that would qualify as "unique". For example, if you use one of the "mood" generators to get a palette of colors that were used harmoniously in lots of maps, does that qualify as using AI to do the map? If you start from a rough sketch and let the system do its thing, how much of the human is there? If you generate a full map by hand and then get some border or field decorations, how about there? If you start with a model, train it on a corpus of your own work and then get back things in your style, how close is that to questionable?
My want has always been (for over 40 years now) for a genie that will let me paint a few areas with magic paint and let the genie do the hard work that requires the kind of detail-oriented talent that I lack. We've gotten closer and closer to that want for years (cloning, Poisson blending, and now "AI"). But it's only in the last few years that I have paused to consider how much I would matter in that process and I'm very much not sure these days.
Personally, if you're using generative, tools, I would like to see information about which tools and which elements on the final map were so generated. I also like to see this information for current works (paint program vs. paint on canvas, stamps from a library vs entirely hand-generated, for example) and I would love to see information on how long it would take folks to get the final result. WIPs are very useful, because (especially for things like paint programs such as Photoshop) how the tool is used is frequently as useful as the final image itself for anyone trying to learn.
I personally think that the use of AI for fantasy mapping is a mixture of being both fun and questionable - on one hand, I think that you can have a lot of fun with AI in that it helps to simplify the generation of basic outlines for landmasses, and it can be a useful tool to start off with for a beginner who might not know how to fumble around with generating clouds in GIMP to get the ideal landmass shapes for instance, or to highlight some of the examples from previous posters about generating more complex icons that someone could put on a map. But on the other hand, I also believe that when you do a whole project with AI, you are not quite putting in the effort yourself, as you have no control over where you would like to place towns, mountainous terrains and rivers etc - when I like to make fantasy maps, I prefer to have some kind of control over where I would like to put the various features that define the landmass onto the map.
Although I am looking forward to seeing some AI tool for fantasy mapping that strikes a perfectly good balance between originality and simplicity.
I’m not going to lie, I’m just flat out against it. Back in the day when artists were first courteously asked to try out the apps (before being ungraciously kicked out with their names and work taken) the idea was that you could build it up with your own art and have it help you build things for yourself, which seemed like the most ethical way to use it.
But that’s definitely not how it was planned to be used from the beginning so it’s sad so many people fell for it.
I’m not trying to be mean and I’m not trying to call anyone out on any of this, but there’s just so many bad sides to it that people are completely overlooking, like the amount of energy it wastes and the impact it takes in the world. Not even mentioning having to watch so many of my art colleagues that have died over the past few years, having all their art fed into it instantly so other people could “keep it going” for them…
I know that in maps/cartography it seems like a much cleaner space and there’s less drama? But if the discussion is true, I guess it’s supposed to hit all the different aspects of what machine learning really does.
There’s so much that could be said about it, but I think it also always comes down to the fact that I always wanna be the one drawing things… if machine learning were to draw for me, I’d just flat out quit. It kills any and all joy for me, so I’ve never touched it.
No idea how to make this look pretty, lol ignore me
Commissions open! https://vgen.co/JulieJubz or https://ko-fi.com/juliejubz/commissions
Personally, I view current AI as a brainstorming partner. Its been a very good sounding board for world building ideas. It can bring up topics I hadn't considered, and help me flesh out topics by asking me questions. Now, when it comes to mapmaking I just don't think it is good enough. Image generation frequently has weird artifacts or nonsensical parts. I've experimented with generating floor plans for buildings and to say I wasn't impressed would be a large understatement.
AI is a tool
...and that's all it will ever be. As a Computer Scientist with a PhD and having written a reinforcing machine learner, I can say that with some confidence. It may be able to do some "cool things" and help you consider some esoteric ideas. They can all be summed up as predefined category selectors (via the machine learning process) with a value scoring algorithm they attempt to "maximize" by the feedback it receives. As such, they will never replace you. Don't let the Hollywood ideology supplant actual reality.
The AI biases are very real, but are not actual human biases. They are based on the data and failures of the data scientist to understand the underlying algorithm the AI uses along with failing to understand how a feature set will produce responses and the very human bias in the training sets. This is truly a traditional Garbage In, Garbage Out issue.
So use it to help you do your work just like any tool in your tool chest, but keep in mind that so called "smart" technology doesn't have the wherewithal to even "know" what smart really means. The first rule of AI: There is no actual I in AI (but you may be fooled by it).
I’m not sure if this makes sense, but I've heard that the computing power required for some AI tasks is so massive that we'll probably always be able to tell when something is created by an AI. For example, generative AI like Midjourney can create some interesting images, but they still often struggle with drawing hands. Even if we could fix this problem, AI might end up screwing with something else... if we really need something to look human, it would be better to have an actual human to do that.
I love AI but for some reason it kinda me believe in the concept of a "human soul". Because any AI art you see has none of it, even when it looks really good.
Posting Permissions
Reply With Quote
