The AI Generated Art Thread

Relevant to this discussion, this just poured up in my news feed.

There’s now a market for prompts. The interview with the guy is pretty cool, showing just how in-depth you can go with your prompts, what some of us were talking about earlier. It’s a lot more than “robot with a fiddle”, heh.

I guess I don’t have a very good understanding of copyright law. I thought I could make a drawing of Darth Vader or take a photo of someone dressed as Darth Vader and put it out under a Creative Commons license as long as it wasn’t derivative of an existing work. (For example, like this one.)

Edit: but trying to sell something with such a photo or illustration might get me in trouble with trademark law.

Not sure about Darth Vader himself, but if you make your own villain named “Lord Darkness” in a mask that’s different from Darth’s and a cape that has evil-looking ruffles on it, I think you’re in the clear to do whatever you want with it. You’d be no more derivative than George Lucas was himself when he derived Darth Vader from Jack Kirby’s Darkseid, a dark lord with a helmet who seeks mastery of “the Source,” which is the energy that flows behind all things, and who is secretly the father of the main good warrior who fights against him.

Indeed, if being derivative at that level violated copyright law, there would be next to no science fiction or fantasy published at all.

Now, if you made just some generic pictures of stormtroopers with their helmets off looking moodily into the camera, then you’d be in trouble. And your art would be really boring, too, though many people like boring art as long as it’s “realistic.” (This is particularly common among sci-fi and fantasy fans, who love their cliches almost as much as anime fans.)

Does that mean that the guy who took this photo and released it under a CC license is in trouble, and only exists at the sufferance of Disney? I find it hard to believe, but I guess you do hear about IP holders going after game mods and stuff, though I think usually the argument there is that the mod could potentially have a monetary impact on future games released under that IP.

I was just dissing the obnoxious dude upthread. hee hee

I really like how MidJourney let’s you see the prompts that other people used to create their works. I did manage to get temporarily banned immediately after buying a month for committing 5 consecutive TOS violations by accident. They really don’t like words like rot, blood, flesh, pus.



The bottom picture looks like something out of Shadow of Chernobyl.

Hmmm, it never occurred to me that gross-out gore words would be banned.

I agree about seeing others’ prompts, though. Some dudes are typing in five whole lines. But then again I typed in “green skin huge ears” and look at what I got:

I have switched from Midjourney to Stable Diffusion standalone. Costs nothing and I can generate as many images as want. Results are of course not quiet as good. Has anyone good suggestion for a stand alone batch image upscaler?

There’s now an ARM Mac native build of Stable Diffusion (called Diffusion Bee) that you can download / unpack / run directly without having to manually install dependencies, then build, etc.

I don’t have any suggestions, but are you using the GUI that comes with RealERGSAN built in?

Yes, though it scales all. I need one where I can batch upscale a directory. So I can create say 50 Iterations of a prompt and then just upscale the best. Probably too dense to figure out how to use it standalone.

I knew I would regret getting an 8GB Macboook Air! It’s great having it local, but man it runs a lot slower than the web version I was using before.

Hm, the version I was using (on colab, not local) had a separate tab where you would upload a single picture to upscale. Wouldn’t work for a whole directory, but does for “pick the best of 50”.

I wonder what the legality of these images is. Like, could you use them in a game for example? If you work on it enough, you could do a visual novel using nothing but AI images.

Can you sell it for profit then? Can you even use the images in any sort of work, a clippings book, a slightly altered painting (or none altered), a presentation of AI art on an art gallery?

Depends on the AI. Midjourney says do what you want with it.

This is what Midjourney says about it.

You basically own all Assets you create using Midjourney’s image generation and chat services. This does not apply if you fall under the exceptions below. See Terms of Service for full details.

I find it pretty interesting though, if you try to make something with for instance Homer Simpson, it doesnt work - Same with Goofy, or Donald Duck. I guess they have blocked certain trademarks like Disneys, for good reason.

Here is a few examples of when I tried to use Donald duck and Homer Simpson


Also there’s (US) precedent saying these images cannot be copyrighted (the US office has rejected copyright applications) because they have not been created by a human. So you can do what you want with them so far, regardless of what platforms claim.

The current legal standing is:

  • Apparently they are not copyrightable so you can use them as you want.
  • The platforms can limit the use of their software, but not the copyright of the images themselves (liability is minor, very hard to claim damages, they can at most ask for some “subscription” money and/or cut you from the tool).

Things that will likely change soon, because it makes little sense:

  • They will be copyrightable.
  • Platforms will be able to give you the copyright (will depend on platform, subscription tier, etc…).

Things that are unclear as of now:

  • Copyright holders of images used for training these might have a copyright claim and thus make all existing systems unusable. I can’t really comment on the likelihood of this, it’s going to depend heavily on how the first cases are managed, and if copyright law is relaxed beforehand or not.

Well, this is a thing now

https://www.youtube.com/watch?v=ZCKdPhepB1s