This is an archive of past discussions. Do not edit the contents of this page. If you wish to start a new discussion or revive an old one, please do so on the current talk page.
The following discussion is closed. Please do not modify it. Subsequent comments should be made on the appropriate discussion page. No further edits should be made to this discussion.
Commons:AI-generated media, which was created in 2022, had a lot of debate and refinement last year, but has been completely stable this year and seems to have calcified into something that more or less reflects the community consensus on how to handle AI-generated media. It's been cited in dozens of deletion discussions as well as user talk page and village pump discussions. Should it be made an official guideline? Nosferattus (talk) 19:23, 24 May 2024 (UTC)
Mostly looks good. A few concerns, though: the scope question isn't actually answered; the implications of some of the legal points aren't explained; and AI-modified images are included alongside AI-generated images. For the latter, there are a variety of denoising, sharpening, scaling, "enhancing", colorizing, etc. applications, some of which are AI-based and some not, but with similar effects. e.g. Photoshop has some new AI tools to replace selected areas, which is just a somewhat better version of a non-AI feature called "content aware fill" that's been part of the software for years. We should probably be documenting when a lot of those are used, but in general if it's not an image that's generated by an AI, but rather the modification of existing images, I probably wouldn't include it in this guideline. Otherwise I'd rename the guideline and add a clear section about AI-based, or non-AI-based but powerful editing tools that do the same thing, software used to modify existing images as opposed to the generating new images. — Rhododendritestalk | 14:57, 26 May 2024 (UTC)
At the end of the day there should probably two different guidelines/pages for AI generated images versus AI/normally enhanced ones. Although I think at this point both AI enhanced and AI generated issues have a lot of the same issues, but there probably needs to be separate guidelines for both regardless. --Adamant1 (talk) 16:31, 26 May 2024 (UTC)
The more aggressive end of AI upscaling is very much generating (and occasionally hallucinating) new content that wasn't there before, when the starting image is of low quality or resolution and the software is being asked to serve back a much higher quality version. Something like File:Napoleone Ratti.jpg (derived from the very low quality historical photo File:Napoleone Ratti original.jpg) is basically asking AI to generate an original human face - but using the outlines of a blurry photo as the constrained input, rather than a detailed text prompt. Belbury (talk) 15:07, 4 June 2024 (UTC)
@Rhododendrites: I have removed the small amount of content related to AI-modified media so that the guidelines focus solely on AI-generated media. Could you clarify what specific "implications of some of the legal points aren't explained"? Thanks! Nosferattus (talk) 19:49, 3 June 2024 (UTC)
Thanks. I went ahead and removed another piece related to AI-modified rather than AI-generated content (and a few other minor edits). Hope that's alright. Support — Rhododendritestalk | 21:11, 3 June 2024 (UTC)
Support The European Union passed a law regulating AI. This project should pass the proposed guideline as well, especially for the good of this project. Well, I realize that jurisdictions vary when reading the guideline. Furthermore, certain cases are subjective. The guideline might change substantially... or not? George Ho (talk) 08:08, 31 May 2024 (UTC)
Neutral. While I certainly feel that a policy on AI-generated and retouched content on Commons is necessary, it's not clear to me that this document provides any substantial guidelines yet - as it currently stands, it's primarily focused on the legality of AI-generated content, not whether it's appropriate for Commons. Omphalographer (talk) 21:39, 2 June 2024 (UTC)
The only one that concerns me is you are expected to document the prompt used to generate the media in the file description. I'm not sure that is, strictly speaking, necessary to have freely-licensed content and may be something that users may not want to document. Not sure why we would require that in general, though maybe could be proof of authorship for previously-published works to avoid VRT. Most everything else seems OK. Carl Lindberg (talk) 00:43, 5 June 2024 (UTC)
Agreed. This guideline doesn't make a lot of sense anymore, given that many popular hosted image generators (e.g. ChatGPT, Bing Image Creator, Midjourney, etc) don't use a textual prompt in a way that's visible to the user, and are frequently updated in ways that would make results non-reproducible anyway. Omphalographer (talk) 02:40, 5 June 2024 (UTC)
I believe I was the first to put forward the proposal that prompts be documented; if not, I was one of the first, and I still stand by it. It was never about reproducibility. It was/is about two things: (1) if an AI image is being put forward as representing something in particular, it seems important to know whether that was present in the way the AI was prompted, vs. someone saying after the fact "hey, that looks kind of like such-and such." (2) There may be some historical value in knowing what might have been a typical output for a given generative AI at some point in its (presumably ongoing) evolution. Frankly, for most outputs of generative AI, that documentation of the AI itself is about the only reason they should be of any educational value at at all. - Jmabel ! talk05:05, 5 June 2024 (UTC)
They would certainly be helpful, and maybe could be encouraged, but this seems to say they are required, and seems like it would generate DRs on that basis alone. Changing "expected" to "encouraged" would be fine for me. Carl Lindberg (talk) 10:16, 6 June 2024 (UTC)
Even if you can't make a 100% accurate reproduction of an image with a prompt they are an important way to know if it's potentially a derivative work or not. --Adamant1 (talk) 05:53, 5 June 2024 (UTC)
Striking Oppose.0xCollection Dvorak DreamsMuch or most of these proposed guidelines such as How should AI-generated media be handled? seem to be written entirely for specifically user-generated content created with prompt-based text-to-image generators of the past couple years, rather than for AI-generated media in general. At the very least, this would need to add clarifications that this proposed guideline is for the subset of AI-generated media that is created through prompt-based software. For example, I am including here a photo of AI-generated media by Refik Anadol. For this type of image of AI-generated media or anything similar, it doesn't seem helpful to have this guideline saying "give the name of the specific AI engine used, followed by the name of the person who created the prompt ... you are expected to document the prompt used to generate the media in the file description." Elspea756 (talk) 17:13, 5 June 2024 (UTC)
You'd still want to know what software was used, and if no prompting was involved, that's fine, say that. - Jmabel ! talk17:57, 5 June 2024 (UTC)
I have no idea what software was used for that work, and I also have no idea whether any prompting was used or not in creating that work, or who would have created any prompt(s) if they were used. I suppose anyone who thinks this a good proposed guideline as it currently is should then feel free to follow the proposed guideline and provide "the specific AI engine," "prompt used for that image," and "name of the person who created the prompt" for that image. I have added it to my watchlist, so I will see any of your updates as you follow this guideline. Elspea756 (talk) 18:42, 5 June 2024 (UTC)
I am not interested in any of this information, but, yes, anyone interested in the "the specific AI engine," "prompt used for that image," and "name of the person who created the prompt" being called for in these proposed guidelines would likely have to try to contact those three. I have already added the image to my watchlist, so I will see any of your updates as you get responses from them as you follow this proposed guideline. Elspea756 (talk) 19:11, 5 June 2024 (UTC)
The fact that there may be a few AI images where this cannot be determined is not much different that the fact that we have some anonymous older artwork, or things that are clearly in the PD for which we don't have clear sourcing. Those don't change the general requirement that images should indicate author and source. -Jmabel ! talk05:13, 6 June 2024 (UTC)
Yes, "images should indicate author and source" is a guideline I would support. The proposed guideline of "the specific AI engine," "prompt used for that image," and "name of the person who created the prompt" is not something I support. Elspea756 (talk) 13:43, 6 June 2024 (UTC)
If we know that little about it, how do we know there is no human authorship, making it a copyrightable work and the photo therefore a derivative work that needs someone else's permission? Carl Lindberg (talk) 10:22, 6 June 2024 (UTC)
The uploader says "I, the copyright holder of this work, hereby publish it under ... the Creative Commons CC0 1.0 Universal Public Domain Dedication." I don't think the uploader following the proposed guideline and also listing a "specific AI engine," "prompt used for that image," and "name of the person who created the prompt" would make any real difference. Elspea756 (talk) 13:49, 6 June 2024 (UTC)
That's fine for the photograph, but what about the pictured artwork? What is the license on that? If copyrightable, the photograph is a derivative work. Carl Lindberg (talk) 01:29, 7 June 2024 (UTC)
At the end of the day guidelines are just that, guides. They aren't policies. But its good to say it would be worth including the prompts then not, because I can guarantee most people don't or won't think about it to begin with. No one is calling for anyone who doesn't include a prompt to be blocked or anything though. So this whole side thing just comes off as super nitpicking. Since when are we not allowed to make suggestions about best practices or guide people on the best way to do something? Because that's literally all this is. --Adamant1 (talk) 02:08, 7 June 2024 (UTC)
If this is written entirely as a guideline for specifically user-generated content created with prompt-based generators, rather than for AI-generated media in general, I think that should be clear, rather than being described as for all "AI-generated media." Elspea756 (talk) 03:09, 7 June 2024 (UTC)
I don't even necessarily disagree, but then that alone isn't a valid excuse to hold up the whole thing either. It's not like the few minor issues can't be resolved once it's implemented. Otherwise it's just bikesheding. --Adamant1 (talk) 09:58, 7 June 2024 (UTC)
I get the feeling you think that's of gotcha or admission of something about this on my part. Regardless, I'd totally support improving the guideline after its implemented of course. That's usually how it works with guidelines. Proposals are just that though, proposals. They aren't meant to be 100% accurate documention of every single possible edge case or over exaggerated scenario and you can't really improve a guideline that exist to begin with. So...I'm all for improving things once they are implemented though. --Adamant1 (talk) 12:54, 7 June 2024 (UTC)
No, that is not a "gotcha" at all. That is a sincere "thank you" for saying you don't disagree with the changes I've suggested to the proposed guideline. I wasn't not expecting so much disagreement from others to what I thought was a very reasonable, clear, and simple suggestion for clarifying the scope of much of the proposed guideline. I am currently making some of those changes to the proposed guideline so that I can hopefully support this proposal. Thank you again. Elspea756 (talk) 13:03, 7 June 2024 (UTC)
Uh OK. Thanks for clarifying what you meant. I guess I just always assumed it was exclusively for AI created artwork since the guideline is literally called "AI-generated media", not "AI enhanced media" or whatever. It would probably be to everyone's benefit to just create Commons:AI-enchanced media as a side thing to main guideline specifically for that aspect of AI software though. --Adamant1 (talk) 13:12, 7 June 2024 (UTC)
To be clear, yes, I am talking about "AI-created" media. Not all AI-created artwork is created with a system that uses prompts, but this proposed guideline was largely written as if that were the case. I've made the sort of edits that I'd proposed in my original post here, making clear that certain sections are only for the subset of AI-generated media that is created through prompt-based software, and clarifying in other sections that the proposed guidelines are referring to all AI-generated media, not just that created through systems that use prompts. These changes can be seen here.[1] I will wait maybe 24 hours to see if these changes are accepted, or wait to hear from any of the people above who opposed my suggestions, and if my changes are accepted, then I will switch to "support." Thank you again, Adamant1, for not opposing my proposed changes. Elspea756 (talk) 13:31, 7 June 2024 (UTC)
I think it's inherent to the thing that if someone can't provide a prompt for whatever reason that they don't have to. Just like it's a guideline that people provide the original author of something but there's still exceptions for anonymous works. Although I can't think of an AI image generator that doesn't involve some kind of prompting to begin with. But then it doesn't really matter, because no one is being forced by this guideline to provide a prompt if there isn't one to begin with. --Adamant1 (talk) 13:39, 7 June 2024 (UTC)
Support Yes, with the clarifications I've made, that Adamant1 said they don't necessarily disagree with, that have survived the weekend, yes, I can now support this proposed guideline. Elspea756 (talk) 23:39, 10 June 2024 (UTC)
@Elspea756: Yes I have, including the section Copyright of AI-generated media modified by humans. If the image was entirely AI-generated it's fine, but if there were significant human alterations later, it's not (the additional human contribution can easily have a copyright). If we know nothing about the image, it's dangerous to assume how it was made. Carl Lindberg (talk) 13:46, 7 June 2024 (UTC)
There is a bit of a stretch between "we need all the detail" by Carl Lindberg and "Nah, it`ll be fine" by Adamant1 and associates. Having looked at most of the stuff uploaded for one week, I tend to go with we need all the info. Alexpl (talk) 14:05, 7 June 2024 (UTC)
I wouldn't say that's my position. I'd just like see a proposal get approved for once instead of getting dragged down by pedantry. Although I have yet to see a legitimate argument for why prompts should be required, not just suggested. Let alone why AI enhanced artwork should be included in this. So I'm more then willing to throw chum to the people who think they don't matter as a compromise so this doesn't just get derailed over something that probably doesn't really matter in the grand scheme of things, if at least at this point, anyway.--Adamant1 (talk) 17:45, 7 June 2024 (UTC)
The discussion above is closed. Please do not modify it. Subsequent comments should be made on the appropriate discussion page. No further edits should be made to this discussion.
Support. — 🇺🇦Jeff G. ツ please ping or talk to me🇺🇦 10:20, 25 May 2024 (UTC)
Oppose
Neutral. While I certainly feel that a policy on AI-generated and retouched content on Commons is necessary, it's not clear to me that this document provides any substantial guidelines yet - as it currently stands, it's primarily focused on the legality of AI-generated content, not whether it's appropriate for Commons. Omphalographer (talk) 21:39, 2 June 2024 (UTC)
