I've heard a picture's worth a thousand words. So, here's a screencap of me in action today:

OK, so it's a screenshot. I'll give it more context, for those of you who haven't been following the posts here in detail in a moment.
But first, let's deal with the confusion I'm reading in the commentary above. I'm afraid I wasn't clear enough, and how I wrote the policy update may have muddied the water, and I apologize if that's the case.
If somebody makes a 1:1 copy of your ships, that's
clearly copyright infringement, regardless of the source. If it's a kitbash that's obviously a kitbash, ditto. Those rules have been in place for years, and AI stuff isn't going to change those rules.
I'm not excusing or condoning
any form of copyright violations. Using AI to make it doesn't change the copyright status of
your work, obviously. So, if anybody infringes on your copyrights... I'd be first in line to take it down from here. I hope that's a satisfactory answer on that point of contention, and again, I apologize for the misunderstanding that appears to have happened, due to me not wording things carefully enough.
Lastly, I've updated the policy, to make everything I've said here clear. If you think it can still be improved for clarity, please PM me; your feedback
will be taken seriously. But don't clog the thread here, please.
Now let's talk about what you can see in that screencap above. What you're seeing there is a screencap of my browser, while I used Easy Diffusion to make one of the two ships in this post.
See the Model field? It's just the current version of
Dread. I've already discussed that model in the
detailed directions I gave out about how you can get started with AI. It isn't trained on anything remotely like Starsector's art, so far as I know. Its creator is into horror imagery. I just like how it produces muted colors, gritty details, rust, stains and biomechanical references.
All the settings, etc. are visible, too. You can clearly see the source image. It's not one of your ships or anybody else's here, either now or from the past. It's an original design I made in 10 minutes in Photoshop.
So, let's go over what facts you've been given:
1. You've all been taught how to make your own art this way. You can- quite literally- see for yourselves that
everything I'm saying here is true by using the technology in exactly the same way, yourselves. There's no "trust me, brah" involved; you just have to go set up the software and follow my detailed directions and examples.
2. You know, with as much certainty as I can provide, that I'm using this stuff, exactly as I've described. IDK how else to "prove" that to strangers on the Internet, but at least I cared enough to try?
3. You know it's legal, and not considered infringement, because the tech doesn't just copy things; it
attempts to recognize patterns, which are then
reinforced by subsequent passes.
That part's pretty complicated to explain to non-techies, but basically... when it's used the way I use it, it's trying to guess what I'd like the results to be...
via inference. Basically, the AI looks at the source image and tries to infer what it's seeing. "Is that a wing? I think that's a wing; let's make it look more like a wing" and so forth. It does that a few million times and spits out a result that, hopefully, looks like a spacecraft.
4. In the screencap, you can see the source image- it's an original design, made by me.
5. I've tried giving a non-technical answer about why I can't train a model to copy your work, even if I wanted to. Here's a slightly-more technical version.
Basically, it's because there isn't enough artwork here to
generate inference well. Stable Diffusion's main model, which every other model gets merged with, was trained on
billions of images. It still can't quite figure out human hands, lol.
Dread, according to its author, uses
30,000 images to train with... and it does an OK job, sometimes, at what it was originally meant for, which is certain types of horror imagery. That some of what it does works well for my purposes is just luck, really- I tried a bunch of models out, and kept returning to Dread, because it works for the things I'm doing with it (which largely aren't SS-related, but that's neither here nor there).
Generating inference is a
hugely complex computational task (like, starter-kit for doing this is an
A100), and for it to work on visual art stuff, you generally need at least
thousands of images of the type you want the AI to infer from. There aren't that many ship sprites in every single mod on this Forum from the last decade (and they're all sorts of different styles, to boot). It just wouldn't work. Not "wouldn't be great" but "would be so awful that it would be worse than useless". We can, via good ol' Occam's Razor, safely conclude that I'm not doing that... and wouldn't bother, to be really honest, because that sounds like work, just to... er... give away free spaceship sprites for anybody to use? Who does that? Some Evil Genius, with an infinite budget... wastes their time on giving away freebies? It's not terribly likely, is it?
So... go look at the tutorial and try it out. Don't take my word for it- I'm not asking for trust. Use it and try to make a spaceship sprite; you've got all the knowledge
right here, in a few Forum posts. You'll realize super-quickly that this tech doesn't magically copy some poor artist's work; you can't just say a few magic words to the AI and get a cool, pixel-art spacecraft.
If you have questions about using it,
ask me via PM. I'm willing to discuss this stuff with you- any of you, on any question you like- but I
do expect that the posts here that aren't about giving away free art to the community will stop, as of now. If that doesn't happen, I'll have to report your behavior to Moderation, because this is really starting to feel like harassment. I don't like doing that. I think that it's better to have a polite, adult conversation in private, and not mess with this thread. Please understand that this really is the last time I'm going to warn you that this behavior is pretty clearly in violation of Forum policy, and my own desire to continue to provide this service to the community is largely dependent on whether I feel welcome here.
AI-generated art is clearly an important technology. But, in the end, it's merely an
accelerator. You still have to have good ideas, and in the case of SS sprites, you need to do final pixel-art work, or the results will be meh at best. I strongly encourage everybody to experiment with the technology; it's useful and fun, and can save time. But it's not a magical copying-machine; it takes some skills to use, and other skills to get a polished final result, and it's simply impractical to copy your stuff, because the tech just won't do that.