AI map development?

[- JO -]

I love the harbour inside the city which stand at least 20m above the sea level.... !

Which AI did you use ?

[Redrobes]

These are all really good and illustrative of the state of AI in maps. The top down room is a good example where initially it looks great. But then you see the walls up against each other. Then no doors. Then you look at the stuff on the floor and I wonder what are they exactly. Then looking at practically everything in all of the rooms they have the right kind of approximate shapes of things but you just cant exactly say what almost anything is in it. You have what appears to be windows but then they look like bookcases. And the list goes on. Superficially is looks alright. The art is great but it just doesn't make any sense whatsoever.

The hut has steps going up with a nice hand rail. Only there is no door at the end of it. There is a collection of rectangles with approximately the right perspective angles but what are they. It doesn't make any sense again.

The second image is also interesting in that its complicated but the town doesn't make sense. But more than this, whilst most of the shadows point in approximately the right direction you note that no house shadows any other house. There is no self shadow. The AI doesn't understand that its a 3D object and has a global view of the town. Its only placing bits of town down in a certain style. The effects of one part of the image is not being projected correctly to another part of it because AI doesn't have any clue about shape. It just renders patches of art.

On all of the images the shadowing is wrong in most cases. It appears to cast a reasonable shadow on the ground but never onto any object in the scene.

In all of these cases the art is not bad but the technical side is utter bonkers. Picking AI stuff out when its pure art is hard. Pure artists have a problem with AI stealing their jobs. But with anything involving a technical aspect to it then its way off. Its clear that all of these images are better than we could possibly have imagined it being from a couple of years ago. The pace they are getting better is astounding. But right now your pretty safe with cartography.

[Azélor]

Which AI did you use ?

Stable diffusion. By default, it's kinda inferior to Midjourney but the later is more limited. Midjourney can give decent output with limited effort but has limited tools to refine/tweak the image. Also, and more importantly, you have to pay and post everything on a public discord. I really hate that.
Stable diffusion has much more tools to tweak the results and you can run it locally on your computer for free.
Of course, you could generate an image in Midjourney and then improve it in stable diffusion or Photoshop, or any other software.

But right now your pretty safe with cartography.

and if you are into hands. The AI really struggle with those.

[Naima]

Honestly I would like to find a free tool like adobe firefly , tgats much more usefull and gives more control, but unfortunately Adobe is greedy and puts paywalls on every feature... but whats annoy me the most is that asks rent rather than buy options.

[Azélor]

Adobe just took tools that already existed for months for free and locked them behind a paywall.
The only reason to use it, is because it's more convenient. Or if your computer has a mediocre graphics card.

Stable diffusion has plugin to use it with Photoshop like this one, but I never tried them.
I'm using an old version that is not supported.

Controlnet alone blows Adobe Firefly and Midjourney out of the water.

[Naima]

Adobe just took tools that already existed for months for free and locked them behind a paywall.
The only reason to use it, is because it's more convenient. Or if your computer has a mediocre graphics card.

Stable diffusion has plugin to use it with Photoshop like this one, but I never tried them.
I'm using an old version that is not supported.

Controlnet alone blows Adobe Firefly and Midjourney out of the water.

Problem is I have no idea how to install this stuff ... What I liked was the lazo tool I could encircle the area I wanted and it created a contextural generation.

[Wingshaw]

I'm in the camp of "I'm fine with AI as a tool to help artists, but I'm against it scraping artists' work and using it without consent". For that reason, I've enjoyed playing with it a bit, but would never share something I've made with it (commercially or not). Once they start using licensed artworks, I'll be more open to the tools' usefulness.

That said, the main reason I decided to comment is that, while I agree with Redrobes' analysis about the flaws of the tech at the moment, it can still be a very powerful (and ethical) starting point for a map. For example, the top-down room has all the problems of detail and technical features that Redrobes mentioned; but one could still use that as a layout generator, a colour palette generator, and an inspiration. The cartographer can then trace over the AI image, and render things more normally. Which also means the artist is still creating a new piece of artwork, not at the expense of another artist.

Just my two cents

Wingshaw

(PS: Hi folks! I've been absent for a while, but lurking in the shadows )