AI map development?

[waldronate]

The ethics of using LLMs depend entirely on the LLM and the data set it was trained on. The early visible LLMs like DALL-E were trained on data sets with no licensing information (or that ignored licensing terms). They generate nice results that freely use license works because that was part of their training data. Other models are trained only on properly licensed data. The current Adobe model, for example, is trained on the licensed images from their broad selection of stock imagery and also from every image that customers save to their conveniently-provided-as-default cloud storage (if you store it on Adobe's cloud, you license it for use in training their model unless you have specifically opted out or avoided their service).
The fun part of LLMs (whether they're attached to a graphical generative part or not) is that they reflect their training data perfectly. That's why so many of them quickly become rabidly xenophobic after a few iterations and so many of them are extremely rude without serious hidden constraints. The internet is like that. Because the LLM has no sense of morals, it just repeats what it was taught. On the internet, that's usually stuff devoid of what most of us like to consider as the "good stuff" in human interactions.

[Azélor]

AI image generation is a tool.
Like other tools, if you don't know how to use it, you can't get great results.

Generating good maps with Ai is almost impossible at the moment. The details and consistency required are out of reach.
Also, since most maps are quite large. Ai can't handle very large image. Upscaling can do miracles but is not that useful on maps.
But it's great to generate map elements such as icon or brushes.

So far, Dall E 3, which what just released, has the best image composition and it the closest thing to an Ai map generator.

[selden]

The ethics issue has been argued on both sides ad nauseam. One argument that I personally find persuasive is that that's how we all start to learn to draw things: by looking at what other people have done and doing something similar.

[selden]

I'm just a novice at prompt creation, but I tend to agree with Azelor. For example, I've found that the free DALL-E 3 provided by bing/create tends to be much better at interpreting prompts than the free version of Stable Diffusion XL provided by NightCafe. Despite my trying lots of different workarounds, SDXL randomly ignores different parts of my prompts, which can be quite frustrating.

I did finally get some (sort-of) flat maps from DALL-E by including the qualifier "line-drawing", but the results still were not very good.

[Naima]

What about that?

https://www.google.com/amp/s/www.tec...rative-ai/amp/

[waldronate]

Nightshade will be the tool that paves the way to an AI Dadaist paradise!

[Diamond]

I've used AI to come up with decorative border elements and some city icons, banners, etc. I don't see anything wrong with using it in such a way - like a lot of other stuff, moderation is key.

[XCali]

That is actually quite nice. Good inspiration how to do different architecture icons.

[jshoer]

For me, the two most helpful points to keep in mind about generative AI have been:

1. It's a tool. Like any tool, it was built by humans who made design decisions about it and is used by humans who made decisions about how to apply it. Anything problematic about its training or use -- and there is A LOT that is problematic -- does not come from the AI, it comes from the companies that made and market it and the users who decide how to use it in their work.

2. It's a "seems like" generator. When you prompt an AI, you're not asking for an answer to a question or an image that matches your prompt. You're asking for something that SEEMS LIKE your prompt, based on everything in the AI's data set that has similar labels. Which is why AI will happily generate references to scientific papers that don't exist, or can't put the right number of fingers on humans hands, and so on. But it is fundamentally not coming up with any new ideas to respond to your prompts.

My attitude is, sure, use it as a tool, but know how you're using it, use it responsibly, and don't oversell the results.

[Azélor]

Nightshade might work but I am skeptical.
In regard to stable diffusion in particular...
Large models use enormous database.
For instance the latest stable diffusion model (SDXL) was trained on the Liaon-5B dataset.
It has over 5 billions images. You can check it here in all it's glory.

I searched a few keywords and the results are not impressive.
To summarize, it uses a lot of images but

The quality is inconsistant. Most images are crappy.
The tagging of images is poor, especially compared to an image database like Danbooru (NSFW).

For an AI, the database is dumb as hell. In that regard, I don't know how Nightshade could affect these large models.

But, most people do not use the main models... They use fine tuned models that are retrained on smaller curated dataset. Maybe a few thousands images at most.
It could work but if you carefully select the images, you could somewhat avoid the poisoned ones. Or only use older images.

Thankfully, there are projects like this one that aim to train model on good dataset ethically.
Even with a much smaller dataset, if the quality is much better, it could result in a much better model.

---

Here are some examples of map (map like) images I've generated.
I can't say that I've spent that much time on that topic, but I've generated quite a lot of AI images.

If you are curious, you can just drag these images in Comfyui and you will get the complet workflow.

ComfyUI_12559_.png
ComfyUI_08095_.png
ComfyUI_08019_.png
ComfyUI_07984_.png

These images were generated during the SDXL beta on their website and I'm not sure how to replicate them:

Indie game art,(best quality, stone cottage, ecopunk, paradise, flowers, verdant, tropical, (St.jpeg
a drawing of a floor plan for a base in Rimworld, house illustration, , d & d design, interior .jpeg
a city in Simcity3000, high density, isometric (1).jpeg
chrome_image_1 juill. 2023 20 h 40 min 24 s HAE.png
chrome_image_2 juill. 2023 11 h 55 min 02 s HAE.png