Interesting. How were these made?

Would you be able to provide links to those tools? Especially their Terms of Use.

-Some kind of AI, I just took Linux version 42MB, and some license is here:
GNU Affero General Public License v3.0
https://github.com/hlky/stable-diffusion/blob/main/LICENSE

..and this guys on itch.io are talking about windows 3GB version here, nasty thing, right. 4GB of memory is minimum to work.
https://grisk.itch.io/stable-diffusion-gui

Care to share what prompt you used if you kept record of it? I know people don't always do so though. Also what subscription model did you use to make 200+ of them. I am assuming not Free Google Colab as they do impose a limit daily. Also yes it is OPENRAIL license. https://bigscience.huggingface.co/blog/the-bigscience-rail-license which "it does not impose any restrictions on reuse, distribution, commercialization, adaptation as long as the model is not being applied towards use-cases that have been restricted."

Awesome to know. yeah I usually have problems getting mine to render highquality like professionally done 3D renders.

These look extremely good, I couldn't believe it was made by an AI.

Anyway I'm still personally pretty paranoid about AI generated content, I'm not for banning it here but could we maybe require a special tag for AI? Just for people like me who prefer to avoid AI if possible, purely for fear of legal complications. Maybe in the submission form under "is this your own work" there could be an AI option. There has been a huge boom of these submissions and I think we should maybe have it under a bit of control.

This is wonderful. AI is a game changer.

I agree people shouldn't have to constantly see the AI artwork taking up the Latest Art feed. If not give it's own tab section I believe it might need a section at the bottom of the art feeds. So underneath where the Featured Art section currently is. I think either would be appropriate as it's going to end up flooding in I feel this will be inevitable as it takes more territory. I still like seeing the art though. You can even make some pixelart into 3D looking images.

You know, that is actually a cool idea... Making a section dedicated to AI. While I disagree that AI in the Latest Art section is a problem, you may be onto something, Joshagibby. The only problem is we may want to wait for more than just three of four users submitting AI generated art before we create a new section.

Yes I agree with you waiting is obviously okay. I'm currently trying to turn real photos of objects into "2d-pixel-art" just trying it out with a paintbucket for starters. So far I have been having a tough time generating visually good looking things to appear in general. Takes me so many tries that I get angry about it at times or depressed. Eh well I will keep going back to it anyway. I am also wondering what to type out for "game related objects". That's why I haven't posted anymore AI stuff. Anyways this probably needs a forum topic to discuss things atleast a post about it people can chat back to eachother about generating.

An AI tag or AI art category aside, (not bad ideas, BTW) I will always insist on submitters listing which specific AI technology was used. AI art generation tools have very different Terms of Use, and many are not compatible with the licenses accepted here.

If you see someone submitting AI art and they say "Oh, it's free." when asked how it was made, you can be sure I will arrive weilding my moderator hammer.

@Sir Twist: Thank you for sharing the art and the link to the AI's TOS. Please add the name of the AI tool as well as that same link to the description of the submission. This will help show the licensing is correct and truly free to use by everyone.

Yes @MedicineStorm we all greatly appreciate you're always checking this! Personally I'm just worried about the complex legal situation here that's furthermore still without many precedents. For example a badly designed or badly configured AI may very well create something that's very close to a copy paste of something in its dataset which may potentially not be free, it's extremely hard to check with proprietary online AI tools. Also I don't know who the burden of proof of having used the paid version would be on -- if the AI owner decided to sue, would WE have to prove it was made with the paid version or would they have to prove it was NOT made with the paid version? Would this be the same in other countries than the US? Is it even legit for a company to limit the copyright of something merely created with their tool, as if could a hammer manufacturer claim rights to the design of a house I made with their hammer? Also examining the compatibility of proprietary terms of service with free licenses won't be trivial and is risky. I'm just worried about the whole complexity of this and for this reason would like to stick to reusing the good old "hand made" assets whose legal mechanism are well established etc. This is why I would like to have a special AI category or something. Apart from this I think AI is a great tool that can save a lot of time and can be very useful of course.

For example a badly designed or badly configured AI may very well create something that's very close to a copy paste of something in its dataset which may potentially not be free

The submitter is responsible for ensuring nothing in the submission violates trademark. Besides that, we as a community are pretty good at identifying those issues when the submitter misses them.

As for copyright, the AIs that I have seen so far on OGA are trained on safe datasets or would not have copyright issues in their outputs. As you say, it is possible for an AI to produce output that looks very similar to copyrighted content... but (dis)similarity of two assets does not effect copyright. It does for trademarks, but not copyright.

It is true a reasonable person may think they are the same and therefore copyrighted, but at that point the onus is on the copyright holder to prove there was copying involved, and not just coincidence.

It is reasonble to be concerned over such issues since "legally ok" and "provably ok" are different things. However, in my opinion, such coincidences aren't any more likely to happen in AI art than the same coincidences in hand-crafted art. Just look for "Zelda-style tilesets" on OGA. Some of it does look like copyright problems, but it isn't. There were some that really were copyright problems, but you won't find them because they were removed.

We curate the AI art the same as any other type of art. If there's a trademark issue, it's obvious. If there's a copyright issue, we'll find out about it and fix it because we have a link to the tool, how it operates, and what its TOSs are. :)

if the AI owner decided to sue, would WE have to prove it was made with the paid version or would they have to prove it was NOT made with the paid version?

The AI owner would have to prove it was not made with the paid version. We make a reasonable effort to prevent that eventuality ahead of time. The submitter has to be pretty deceptive to put you in that position. Lies like that get found out and we ban those people. The risk is the same as someone posting non-AI art and saying "uh, yes- YES! I am the author of this art. I didn't steal it at all!" when they definitely did steal it. Avoiding AI art for this reason does not really diminish that risk IMHO.

Would this be the same in other countries than the US?

I don't know much about non-US law (nor am I an attorney) but probably, yes. Civil burden of proof is commonlaw and magna carta and all that: the basis for most countries system of laws these days. Not legal advice, of course, you'll have to check your local laws yourself, I'm afraid.

Is it even legit for a company to limit the copyright of something merely created with their tool, as if could a hammer manufacturer claim rights to the design of a house I made with their hammer?

Probably depends on if you're renting the hammer as a service instead of owning the hammer yourself. It certainly is legal for a company to govern who is allowed to manufacture and distribute their model of hammers, so... good question.

Also examining the compatibility of proprietary terms of service with free licenses won't be trivial and is risky.

Agreed, it has not been trivial so far. Thank you for recognizing my efforts. :P Hahah! You may notice that submissions will be marked with the dreaded "Licensing issue" until I have had a chance to look at the EULA. If the proprietary terms of such services even hint at incompatibility with our licenses, I do not not allow the submission. That is why I always ask for links to TOS on AI submissions. I read through the EULA of every one of them.

I'm just worried about the whole complexity of this and for this reason would like to stick to reusing the good old "hand made" assets whose legal mechanism are well established etc. This is why I would like to have a special AI category or something.

Yes, and I encourage each person to consider their stance on this like you have. For now, I can promise that every known AI submission will mention which AI technology was used so that you can curate it out of your own list of assets you will be using or not. Admittedly not as good as its own category or tag, but I hope that will suffice for now.

OK, very fair points, I'll be still keeping it safe and watching from a distance how this is evolving for now :)

I have been messing a lot with AI just for good old personal fun. I haven't submitted anything that I purposely made directly from something that was already existing. The towers all I did was use a shape of a medieval spire tower with a prompt and you can see directly none of them are anywhere a copy and AI doesn't go off of a particular style type to my knowledge unless you're being too specific to a topic. For example I wanted to make an ogre and when that failed to make it originally I used shreks face. However I then made various shapes around the character. Where I believe focal points would occur based on color & darkness. So I made an ogre after deleting the shrek image. It didn't just make a perfect looking Shrek did he still look like a shrek yes in a lot of ways due to me typing Shrek into the prompt. However I just cut the ears off placed on custom elven ones. Rerendered it then it came out looking like a rather odd custom emoji character face. Expecially since my drawing skills were off on the face shape. Would I post that as something for people to use no despite me liking the results I know that's not proper.

From https://huggingface.co/spaces/CompVis/stable-diffusion-license :

"You are accountable for the Output you generate and its subsequent uses. No use of the output can contravene any provision as stated in the License."

So if I read it right, this seems problematic for redistribution of the output (i.e., generated images) - the "Use Restrictions" don't just apply to the output itself, but subsequent uses of the output. Which seems incompatible with licencing under CC0 (or any licence that would be considered Free/Open), since that licence has no such use restrictions.

Admittedly it seems unlikely that this particular artwork could be used in such a way (hmm, maybe someone makes a medical advice website and wants to illustrate a section on medieval plague...) But more generally, you could have images that seem fine on their own, but could be used in violation of the Use Restrictions.

It doesn't help that the wording is vague (in what way is the author "accountable" if someone takes the images and uses them against the Use Restrictions?) It doesn't explicitly say that Output must be distributed with the Use Restrictions, but it seems worth being wary about.

I also think there's confusion around Stable Diffusion because AFAICT the source is under a standard Open Source licence, but the model is not. Also note, I'm not making any judgement on the pros or cons of the Use Restrictions - just noting that they don't seem to be compatible; it would be the same with any non-AI derivative work that was derived from a non-Free/Open licence.

I believe the license is mentioning the fact it doesn't like the usage of people creating restriction based things that the AI thinks is inappropriate under it's licensing. You know such as generating gore, dead bodies, mutating people, sexually messed up stuff, and controvesial things. In that way I don't see how the license mentions it not being an open license. People who create these can land the people who made the AI model originally in hotwater if they are not specifically clear about the usage beforehand. So they are basically just saying that they are legally exempt from anything that people might generate. While you can not blame them for something messed up being shared online. Hence "You are accountable for the Output you generate and its subsequent uses. No use of the output can contravene any provision as stated in the License."" It's just a safety precation legally for them.

You are solely responsible for determining the appropriateness of using or redistributing the Model, Derivatives of the Model, and the Complementary Material and assume any risks associated with Your exercise of permissions under this License. 10. Limitation of Liability. In no event and under no legal theory, whether in tort (including negligence), contract, or otherwise, unless required by applicable law (such as deliberate and grossly negligent acts) or agreed to in writing, shall any Contributor be liable to You for damages"

At the end of the Huggingface license for it is where it says use restrictions as well.

From https://huggingface.co/spaces/CompVis/stable-diffusion-license

"You are accountable for the Output you generate and its subsequent uses. No use of the output can contravene any provision as stated in the License."

I can't seem to find the quoted paragraph in the link provided. It does say it here: https://huggingface.co/CompVis/stable-diffusion-v-1-4-original in the summary blurb about the license, but not the license text itself.

My interpretation of that was not the same as yours, I feel. The sentence just before the quoted segment says "CompVis claims no rights on the outputs you generate." As in, 'no rights reserved on the output. The model has restrictions, but not the output.' Basically CC0. Again, the 'you are accountable for the output and its subsequent uses' is the same for CC0. If you derive new content from my CC0 content, you are responsible for that derivative and its subsequent uses. As the original creator, I am not responsible for your decision to use my asset in an illegal way.

I feel like the OpenRAIL license summary is saying the same thing that the CC0 summary says, it's just that CC0 phrases it as "the person who associated a work with this deed makes no warranties about the work, and disclaims liability for all uses of the work, to the fullest extent permitted by applicable law."

What did I miss, then? It seems to me like the restrictions being discussed do not place responsibility on the person sharing an asset for other people's downstream usage. Furthermore, it seems like the restrictions being discussed do not apply to the output at all, just the model that generated the output.

Either the license is placing no restrictions on the output, or the license is contradicting itself when it says "claims no rights on the output".

@Joshagibby: If someone wants to licence an output image under say CC0, it no longer has any of those usage restrictions on the subsequent use of that image. I'm not questioning why they've done it that way, just noting that these terms exist.

(N.B., the licence doesn't seem to have any restrictions on use regarding sex or gore; see "Attachment A" in the link.)

@MedicineStorm: It's under Section III, paragraph 6 - in full:

"6. The Output You Generate. Except as set forth herein, Licensor claims no rights in the Output You generate using the Model. You are accountable for the Output you generate and its subsequent uses. No use of the output can contravene any provision as stated in the License."

The wording in the summary blurb seems fine - but the full licence text adds the word "subsequent" (i.e. "subsequent uses" of the output) which is what had me wondering.

Regarding the idea that they're just disclaiming liability, I see what you mean - but it's the way that sentence is followed up with "No use of the output can contravene any provision as stated in the License". Maybe it's just a poorly worded limitation of liability - but it's the way that's followed by that line. I suppose it's true that the line "No use of the output can contravene any provision as stated in the License" no longer contains the word "subsequent". But I feel the licence is unclear.

It's good that it says "Licensor claims no rights in the Output You generate using the Model." which as you say would in itself imply you can licence and redistribute as you like, except the full licence text qualifies that with "Except as set forth herein".

"Furthermore, it seems like the restrictions being discussed do not apply to the output at all, just the model that generated the output."

Paragraph 6 refers to the "output" though.

Essentially the question is "can derivative works under licence A be relicensed as licence B", which is hard when licence A is a relatively new licence.

@marko: that is a good point. I think what you said about it being a new license sums it up. I think this is something of a crayon license and it is difficult to sus out the implications of seemingly contradictory terms here.

With that in mind, what are some ways the end user could use the output in a way that contravenes the model license? I'm trying to think of some use-case that could cause a conflict with the terms placed on the model, but i'm not coming up with much. I mean, yeah, the part about "You can't use the model to deliberately produce nor share illegal or harmful outputs or content." so using output that is illegal or harmful contravene's the provisions, I guess, but that isn't even the applicaiton of output that is conravening the provisions, it's the intentional production of said output that is doing so. And, obviously, "illegal or harmful" outputs are already illegal without being forbidden by that provision. That's true for assets under any license. That's like making a new law that says "It is hereby against the law to break the law".

Parody materials(games, animations) could be seen by some as to disparage/defame groups.

Sorry I forgot to reply to this:

@MedicineStorm "With that in mind, what are some ways the end user could use the output in a way that contravenes the model license?"

I agree it's less likely for the kind of images that would be typically uploaded for use in games. But in theory someone could use AI generated art in a medical application or website, as one example.

But in general, I agree that having people post the terms of the AI tools used to create such art is a good thing.

This submission has been marked as having a potential licensing issue. Hopefully, this is temporary. This does not mean the submitter has done anything wrong or that there is even a licensing issue at all, but there is a potential licensing issue. As mentioned previously in various places on OGA, we said

"...Until that broader discussion [of the legality of AI assisted artwork] progresses closer to some form of conclusion, we'll go with the general consensus. ... [the] general precedent [is currently] that it's fine. If that general consensus changes, so will the decision to host such assets on OGA."

The general consensus is changing now. Changing to what is not yet clear, but there are several ongoing US court cases involving exactly this topic. The court's decision on these cases will inform how OGA hosts assets created with the use of AI assistance. See https://opengameart.org/content/artificial-intelligence-assisted-artwork for more information.

This is NOT any sort of final determination or a permanent flagging of this submission. Regardless of the court's decisions about these tools, there are a lot of factors to take into consideration. Not all submissions will be handled the same way.