Wikipedia talk:No original research

An editor is creating multiple images themselves, uploading them to Commons, and then posting the images to Wikipedia articles. Let me explain the situation and why I am posting about it here, instead of the OR Noticeboard. This editor is fairly new and seem to be doing this interpretive editing in good faith with the images being clearly-marked either as a "digital reconstruction" or as a "digital remake". They are basing their images on written published sources...but. But these images are created interpretations, they are not published elsewhere, they are being published directly onto Wikimedia platforms, in Wikipedia's voice, so yes, this would all seem to be original research. But there is nothing that specifically addresses this issue in the No original research guideline, so it seems to me that some specific wording about this image issue should be added to the Original images section, something along the lines of:

Created images of historical items that are previously unpublished interpretations unavailable elsewhere are original research and even though these images are or might be based on text descriptions in published sources any such images should be removed.

Also, in light of developing technology and AI, this is a type of issue that can come up anywhere within the Wikipedia/Wikimedia platforms and we need to get out ahead of it. Similar images could possibly pollute our historical information streams with created images. I mean, have you seen the "AI suggestions" that are now prominently displayed on the first line of Google Searches? What I've been seeing is that they are often thin paraphrases of Wikipedia articles, so just wait until the "AI Suggestion" bot gets a hold of created interpretations of historical items based only on written descriptions - the images will be repeated and repeated... Anyway, would appreciate any thoughts on adding a line about "created images, even though only based on written descriptions..." etc., etc. Thanks, Shearonink (talk) 06:19, 24 December 2024 (UTC) [reply ]

We've already seen examples of AI images uploaded based on text descriptions. I'd support an addition to the guidance along these lines. Nikkimaria (talk) 14:46, 24 December 2024 (UTC) [reply ]

Nikkimaria I wasn't aware of the AI images, but I suppose is possible in this particular instance that these images are actually AI-generated... Upon re-reading my suggestion above I think it should be/could be re-crafted into:

Created images of historical items that are previously unpublished and are unavailable elsewhere are original research. Even if these images are or might be based on text descriptions in published sources these images are interpretations not actual historical images of these items and any such images should be removed from Wikipedia pages.

Is this new sentence better? - Shearonink (talk) 17:24, 24 December 2024 (UTC) [reply ]

I'd suggest splitting up the issue into two new sentences, one in each paragraph of OI. First: "Original images that are previously unpublished interpretations of text descriptions are considered original research and should not be included." Second would be something specifically relating to AI-generated images, which probably warrants a separate discussion. Nikkimaria (talk) 17:35, 24 December 2024 (UTC) [reply ]

Ah ok, that makes sense. - Shearonink (talk) 18:45, 24 December 2024 (UTC) [reply ]

We might need a community-wide RfC regarding AI-generated images (and even editor-created drawings/sketches/cartoons/paintings, particularly of BLP subjects, in general). I don't believe we have any sort of policy or guidance that addresses these issues. Some1 (talk) 18:01, 24 December 2024 (UTC) [reply ]

Possibly? But (shrug) I do not agree. These images, whether they are editor-created or they are AI-generated...these images are clearly original research, which is against a stated Wikipedia policy and which, seems to me is inextricably connected to the policy of Verifiability. Others may disagree - and that is fine - but I do not think a RfC is needful or warranted. - Shearonink (talk) 18:45, 24 December 2024 (UTC) [reply ]

There was a big discussion about user generated images of people awhile ago, I'm not going to dig around for it now. I don't believe it come to any conclusive consensus, most of the images were removed but I don't believe all of them. If I remember correctly the contention was between having artists impression to improve articles with, and concerns over quality and OR. Take that with some salt, as I wasn't overly impressed with some of the images and so may have a biased memory of it. If no-one else brings it up I'll try and find a link to it once I have a bit more time. -- LCU ActivelyDisinterested «@» °∆t° 18:59, 24 December 2024 (UTC) [reply ]

Thanks - Shearonink (talk) 19:21, 24 December 2024 (UTC) [reply ]

Here are some discussions I'm aware of, not sure if any of them are the one you're thinking of: Wikipedia_talk:Manual_of_Style/Images#WP:USERG_portraits, Wikipedia:Village_pump_(policy)/Archive_190#AI-generated_images, Wikipedia:No_original_research/Noticeboard/Archive_49#Cartoon_portraits. Plus there's WP:AIIMAGE. Nikkimaria (talk) 19:35, 24 December 2024 (UTC) [reply ]

Thanks Nikkimaria, those discussions are very useful - I'm reading through all of them now. Will take me quite a while... - Shearonink (talk) 22:40, 24 December 2024 (UTC) [reply ]

These images, whether they are editor-created or they are AI-generated...these images are clearly original research Oh, I definitely agree with you. I had to revert an amateurish drawing of a BLP subject recently and the editor was quite upset about it. It'd be nice to have something written in policy that directly addresses these issues (AI-generated images/editor-created drawings/sketches/cartoons/paintings, etc). I believe the only place to add something into policy is the Village Pump via RfC. Some1 (talk) 20:09, 24 December 2024 (UTC) [reply ]

Perhaps you are right, that a broad RfC at the Village Pump is needed. I do think it is important to have some sort of policy/guideline in place sooner rather than later...

Also. Personally, I am of the opinion that the guideline against OR should stand, whether or not the OR is text or the OR is an image. If the text is not found in a published reliable source then it simply cannot be used. The same would seem to be true and should be true for an image - after all, they are both content. - Shearonink (talk) 22:40, 24 December 2024 (UTC) [reply ]

I don't see "previously unpublished" as being the right criterion -- surely the publication needs to be in a reliable source. But I'm not sure even that's enough, because in my experience publications that are careful about the text they publish are sometimes a bit loosey-goosey about images. I feel we should allow images such as the lead image in Matthew the APostle, and not editor-created stuff, but I'm not sure exactly where the line should be drawn or how to describe it. E Eng 04:53, 25 December 2024 (UTC) [reply ]

Ok, I see your point. Perhaps this is better:

Original images posted onto Wikipedia pages that are previously unpublished in reliable sources and that are created interpretations of text descriptions are considered original research and should not be included.

Shearonink (talk) 14:57, 25 December 2024 (UTC) [reply ]

Should we start a workshopping section below? I'm thinking something along the lines of:

AI-generated images and user-created artwork (such as original drawings, sketches, paintings, cartoons) posted onto Wikipedia articles that are previously unpublished in reliable sources are considered original research and should not be included.

Some1 (talk) 16:03, 25 December 2024 (UTC) [reply ]

Sounds good. I agree a workshopping section should be started, either here or over on the Village pump:technical. - Shearonink (talk) 17:14, 25 December 2024 (UTC) [reply ]

Try this:

Drawings, sketches, paintings, cartoons, AI-generated images, and so on not previously published in reliable sources are original research and therefore should not be used in articles.

This removes some stuff in the previous proposal that's actually irrelevant (e.g. doesn't matter whether or not it's "posted to Wikipedia" -- it's OR either way). I also removed the bit about "user-created" since if it's not from an RS, it's OR no matter who created it. I considered changing should not to must not, but I'm not absolutely certain there aren't acceptable edge cases. E Eng 16:52, 25 December 2024 (UTC) [reply ]

Yes, you are absolutely right about the OR *but* I think the proscription against "user-generated" images should be explicitly and plainly stated even if it does seem somewhat repetitious to us experienced editors. - Shearonink (talk) 17:14, 25 December 2024 (UTC) [reply ]

OK:

Drawings, sketches, paintings, cartoons, AI-generated images, and so on (whether created by editors or by others) not previously published in reliable sources are original research and therefore should not be used in articles.

E Eng 17:37, 25 December 2024 (UTC) [reply ]

The only further adjustment I might make would be to state "must not" - or to add "must not - instead of "should not". "Should not" implies possible permission, "must not" says no way, don't do it. So maybe something like...

Drawings, sketches, paintings, cartoons, AI-generated images, and other illustrations (whether created by editors or by others) not previously published in reliable sources are original research and therefore must not - and should not - be used in articles.

Shearonink (talk) 18:13, 25 December 2024 (UTC) [reply ]

I mean, the lead image for Gisèle Pelicot is fine, and ran on the front page. File:Light dispersion conceptual waves.gif is fine, and featured. File:Chloralkali membrane.svg is fine, and featured. File:Visit of the Mandelbulb (4K UHD; 50FPS).webm is fine, and featured on enWiki, faWiki, and commons. File:Pi-unrolled-720.gif is a good illustration, featured here and on commons, and a former featured picture of the day. The current draft would exclude high quality and encyclopedic images like these. GreenLipstickLesbian (talk) 07:34, 26 December 2024 (UTC) [reply ]

Was this edit the one where you removed the "amateurish" drawing of a BLP? That image was created by a notable professional artist. If so, that's at least twice this year that an editor has complained about a drawings being "amateurish" when they should be saying something a lot closer to "by a professional artist whose artistic style is not my personal favorite". WhatamIdoing (talk) 19:18, 25 December 2024 (UTC) [reply ]

I didn't realize the person was a "professional artist" when I first removed the image [1] from the infobox, so "amateurish" might not be the right word to use now. Hindsight is 20/20. But my point still stands, that these types of user-created drawings/sketches/cartoons/etc. are subjective and based on what one's own interpretation of what the BLP subject looks like. They shouldn't be used in BLP articles, especially as the lead image. Some1 (talk) 19:28, 25 December 2024 (UTC) [reply ]

I've always been a little wary of this type of WP:OR, or at least WP:SYNTH. But I haven't been sure how to address it, since it is normalized on certain pages. Shooterwalker (talk) 00:54, 25 December 2024 (UTC) [reply ]

Shooterwalker - "It is normalized on certain pages"... Really? Which ones, what subject areas? - Shearonink (talk) 04:18, 25 December 2024 (UTC) [reply ]

I've seen it in articles on fantasy fiction (Hobbit and so on), obscure Norse gods, and stuff like that, where editors think it's OK to make up their own artwork based on textual descriptions. Example: [2]. Such stuff needs to be hunted down and shot on sight. E Eng 04:30, 25 December 2024 (UTC) [reply ]

Yikes. And agree... - Shearonink (talk) 14:52, 25 December 2024 (UTC) [reply ]

@Shearonink There are cases such as this: File:Star_Trek_Timelines.png.

I haven't tried to address it because there are a few highly active fandoms on Wikipedia that basically WP:IAR. But exceptions make bad rules anyway.

More generally, I have seen most other stuff removed and re-organized in accordance with Wikipedia policies. If we can write our policies and guidelines around most cases, we can at least stop things from getting worse. Shooterwalker (talk) 17:43, 2 January 2025 (UTC) [reply ]

I have also started a discussion on WT:BLP about AI-generated images, I was just informed that this discussion also exists so I figured it would be appropriate to share here since it's relevant. Di (they-them) (talk) 23:21, 29 December 2024 (UTC) [reply ]

Another user has started a discussion at the Village Pump after your post: Wikipedia:Village_pump_(policy)#Guideline_against_use_of_AI_images_in_BLPs_and_medical_articles?. Some1 (talk) 13:01, 30 December 2024 (UTC) [reply ]

Re: not previously published in reliable sources

The image to the right (at least on my screen, it's to the right) says that it's an "Original image" and the editor's "Own work. Taken at City Studios in Stockholm, September 29, 2011". The image was not "previously published in reliable sources", but I think everyone here agrees that the image is valuable and usable on Wikipedia articles. How do we write one or two sentences (or even a paragraph) that discourage the use of:

• 1) AI-generated images
• 2) User-created artwork (drawings, paintings, sketches, cartoons, etc.)

that have not been previously published in RS but without being too restrictive? Because unfortunately, I can see proposals like these failing if the community thinks the proposals are too restrictive. Some1 (talk) 18:29, 25 December 2024 (UTC) [reply ]

Why do we need an editor to make up their own image? There aren't appropriate images in (what appears to be) a reliable source like https://openstax.org/details/books/anatomy-and-physiology-2e ? And looking closely, the male is shows as having "breasts", which (correct my if I'm wrong) is not appropriate, and the female appears to be wearing nail polish on her toenails. Both of these are arguments for why we shouldn't have our amateur selves making our own illustrations.Having said that, however, I added the red annotations to this:

and used it in my favorite article, Phineas Gage, with an explanatory caption. I would argue, naturally, that this is OK and not OR, but I'm not sure exactly how to distinguish it from what we've been talking about excluding. E Eng 18:43, 25 December 2024 (UTC) [reply ]

This example is simply the same as adding a caption to a reliably-sourced image. You've added an explanation not altered the image and changed its meaning. - Shearonink (talk) 21:39, 25 December 2024 (UTC) [reply ]

I guess, but I did have to put 2 and 2 together a bit to do it. H marks Dr. Harlow's house, but since the map is 20 years after the events the article discusses (during which time Harlow had moved away), you'll see the house itself is actually labeled "Dr. Hazelton" on the map. Hazelton bought Harlow's practice, and the the location of the "Dr. Hazelton" house is in the same spot as a house labeled "Dr. Harlow" in an earlier, low-quality map I didn't want to use in the article; it also matches texual descriptions in RSs of the location of Harlow's house. So I felt comfortable annotating the map as I did, but strictly speaking one might consider it a step or two toward the OR boundary. E Eng 00:39, 26 December 2024 (UTC) [reply ]

Writing this correctly and concisely is going to be very difficult, even if we agree that it's the right thing to do. WhatamIdoing (talk) 19:15, 25 December 2024 (UTC) [reply ]

Oh yeah, ain't that the truth. - Shearonink (talk) 21:39, 25 December 2024 (UTC) [reply ]

A broad rule covering all AI-generated images and user-created artwork might not pass, but if the proposals were more narrow and focused, e.g. AI-generated images not previously published in RS, or user-created artwork (drawings, cartoons, sketches, paintings, etc.) depicting BLP subjects, etc. I could see them getting support. Some1 (talk) 19:43, 25 December 2024 (UTC) [reply ]

I think the key to user generated images is to make it’s provenance clear in the image caption. Disclaimers such as "AI generated image of X" or "Artistic depiction of Y" resolve most questions. Blueboar (talk) 20:39, 25 December 2024 (UTC) [reply ]

Oh I dunno about that Blueboar...AI-generated or user-generated images that do not exist in reliable sources? Why are these images not OR? In my original post an editor is creating images of historic flags. One example is described in historical texts as a black F on a white field...but no mention of measurements or fonts or the size of the F etc., etc., so what one editor might create will be different from another editor...who's to say which version is more correct than another? We can't. So we shouldn't. - Shearonink (talk) 21:39, 25 December 2024 (UTC) [reply ]

Because the purpose of an image is to illustrate the article, not to convey accurate information. We have long held that our NOR policy does not really apply to images.

That said, NOR does apply to the caption (ie text accompanying the image)... like any other text in our article. Blueboar (talk) 22:26, 25 December 2024 (UTC) [reply ]

...so what you seem to be saying is that anyone can draw up an image at any time and place that into any article because images are mere illustrations? "We have long held that our NOR policy does not really apply to images." Regardless, with editorial consensus, guidelines and policies change around here all the time. - Shearonink (talk) 23:09, 25 December 2024 (UTC) [reply ]

Yes, anyone can create an image and stick it in an article as an illustration.

That doesn’t mean this image will remain in the article. If others think the article would be better with a different image (or even no image at all), simple consensus can determine this. Blueboar (talk) 00:12, 26 December 2024 (UTC) [reply ]

Blueboar, we've long turned a blind eye to a certain amount of OR in images (see my own confession above), but a blind eye it is. With the advent of AI, making it way more tempting to generate ill-founded image trash, I believe the time is coming soon that we need to come to Jesus on this question, and clarify more explicitly what is and isn't acceptable. I think your asserted distinction between "illustration" and "information" is a dodge. Even if an image's caption says "Artist's conception" or "AI-generated image", we are putting WP's imprimatur on that image. Having said that, I'll repeat my statement above the the lead image in Matthew the Apostle is fine with me, even though it implies WP's imprimatur for that image, so clearly I don't have a completely crisp idea of the exact criteria apply. E Eng 00:39, 26 December 2024 (UTC) [reply ]

Depends on what X and Y are. Captions wouldn't help in cases such as: [3]. According to Lundy-Paine's Instagram page, they look nothing like that "portrait". (Anyway, I'm just using that as an example; the cartoon portraits on BLPs problem has already been solved, I believe. But who knows when the issue will arise again.) Some1 (talk) 21:11, 25 December 2024 (UTC) [reply ]

Nothing says we can’t swap one image for an image we think is better. That just takes a simple consensus to achieve. Blueboar (talk) 00:02, 26 December 2024 (UTC) [reply ]

I confess to being flabbergasted by your position here. You're not seriously suggesting that that cartoon image would be acceptable in that article, under any circumstances, are you (other than if, somehow, the article needed to discuss the cartoon itself)? By your reasoning NPOV-violating text, or NOR-violating text, or V-violating text, would be OK in an article since, ya know, someone can someday swap it out in favor of better text. E Eng 01:03, 26 December 2024 (UTC) [reply ]

Sometimes a cartoon is fine (as an example, the Bayeux tapestry is essentially a cartoon and we use it to illustrate our article on Edward the Confessor)... at other times a cartoon wouldn’t be the best choice. I certainly would prefer a more realistic drawing/painting over a cartoon - if one is available... and a photo over that. However, I would happily bow to consensus if other editors thought the cartoon was the best.

My point is that this decision (cartoon vs painting vs photo) is purely an ascetic one, governed by what is available and consensus discussion at the specific article. Our "No Original Research" policy is not an issue. Blueboar (talk) 02:51, 26 December 2024 (UTC) [reply ]

You're simply asserting that NOR isn't an issue. V says that All material in Wikipedia mainspace, including everything in articles, lists, and captions, must be verifiable, and though images themselves aren't listed explicitly I don't see how they can escape being considered part of All material in Wikipedia mainspace. As I've mentioned, a blind eye has been turned to this heretofore, but I think that time is coming to an end. E Eng 04:19, 26 December 2024 (UTC) [reply ]

The Edward the Confessor infobox image[4] doesn't fall into the examples that we've been discussing, considering it is a photograph of the Bayeux Tapestry that was presumably taken in a museum; a regular editor like you or me didn't draw/paint/use AI to generate an image of Edward the Confessor then upload it onto Wikipedia. Some1 (talk) 12:54, 26 December 2024 (UTC) [reply ]

This discussion has me wondering though... what's stopping non-notable artists from uploading their non-notable artwork onto Wikipedia (with their watermarks and all) in an attempt to promote themselves? "[Musician] doesn't have an image? Welp, better draw one really quick and upload it onto Wikipedia to put on their article!" I don't think this phenomenon (with the watermarks) has happened yet on Wikipedia, surprisingly enough. Some1 (talk) 13:19, 26 December 2024 (UTC) [reply ]

Nothing stops non-notable artists from uploading their art. And if the rest of us feel that this art improves an article, it will remain in the article. Conversely, if the rest of us feel that it does not improve the article, nothing stops us from removing that art, or replacing it with different art.

This is how WP works. Things get added, and things get removed... and when there is disagreement over an addition or removal we discuss it and try to reach a consensus. Blueboar (talk) 14:12, 26 December 2024 (UTC) [reply ]

I'll repeat yet again that you could use that kind of reasoning to allow all kinds of policy-violating material to be added to an article. That alone can't be the justification. E Eng 15:49, 26 December 2024 (UTC) [reply ]

You are missing my point... we don’t need to amend a policy in order to justify removing things we don’t think improve an article, and we don’t need amend policy to give us permission to add things we think do improve an article. If an image does not improve an article, we remove/replace it. If an image does improve an article, we keep it. If there is disagreement about a specific image, we discuss it. It really is that simple. Blueboar (talk) 19:37, 26 December 2024 (UTC) [reply ]

While this is technically true, it's also very helpful to have something to point to other than I just don't like it when explaining why you don't think a particular addition is an improvement, and in some discussions about this I've seen the current OI wording being pointed to as meaning that user-created images of any kind are a-okay (example). Nikkimaria (talk) 19:46, 26 December 2024 (UTC) [reply ]

Men have breasts. They aren't gender or sex specific. GreenLipstickLesbian (talk) 07:08, 26 December 2024 (UTC) [reply ]

OK, but how about the toenail polish -- is that part of female anatomy? E Eng 15:49, 26 December 2024 (UTC) [reply ]

Yes, of course. GreenLipstickLesbian (talk) 21:10, 26 December 2024 (UTC) [reply ]

Of course, if anyone wants to find and upload better, or even just different, images in the Standard anatomical position, please feel free. I remember when those photos happened. It was a years-long process that ultimately involved hiring professional models. The modelling agency had a lot of trouble finding anyone who met our criteria (e.g., normal-ish body weight, not heavily tattooed, without heavy tan lines) and was willing to do it. We didn't get everything we wanted (e.g., natural body hair, absence of nail polish), and we were only able to get one woman and one man, but there was nothing else available back then, and this was a substantial improvement. WhatamIdoing (talk) 05:30, 27 December 2024 (UTC) [reply ]

There are all sorts of highly educational editor-created diagrams that we really don't want to lose including diagrams of medical, biological and chemical structures, physics explanations, historical maps, and many more. The vast majority couldn't be replaced with any (recent) published image due to copyright restrictions. Provided that such diagrams are based on written reliable sources, are factually accurate, educationally useful, and agreed by consensus they should be encouraged, not prohibited. MichaelMaggs (talk) 14:44, 26 December 2024 (UTC) [reply ]

Yes, just so long as we can agree that diagrams are completely different from cartoons of living people. E Eng 15:49, 26 December 2024 (UTC) [reply ]

Agreed. User-created cartoons of living people are normally neither factually accurate nor educationally useful. MichaelMaggs (talk) 16:01, 26 December 2024 (UTC) [reply ]

I'm just a tiny bit doubtful about that as an absolute rule. I suggest, for example, that a cartoon for Jaiden Animations would be more accurate, educational, and relevant than a photo of her, since she is largely notable for her autobiographical self-portraits. Of course, in that case, I'd want an authentic self-portrait from the artist herself.

What I'm certain of is that some editors deeply loathe any representation of a person that is not "realistic" in style. Caricatures have to be "accurate" and "realistic" in some sense, else they aren't recognizable. But these editors want something "life like". WhatamIdoing (talk) 05:45, 27 December 2024 (UTC) [reply ]

editors want something "life like" Like an actual photograph of them, preferably, yes. Some1 (talk) 05:49, 27 December 2024 (UTC) [reply ]

• one aspect I'm seeing above to make sure is treated as acceptable are user made images that are recreations of existing published images in copyrighted sources that can be remade in a copyright free version. Commonly this is for graphs from journal articles where the data is available. Masem (t) 16:09, 26 December 2024 (UTC) [reply ]

• I've just learned that there's c:Category:AI-generated images of living people (PIP). Some1 (talk) 15:27, 29 December 2024 (UTC) [reply ]
• The entire user-generated maps genre would be endangered by a misguided policy of banning images based on text. What this conversation is really circling around is banning entire skillsets from contributing to Wikipedia merely because some of us are afraid of AI images and some others of us want to engineer a convenient, half-baked, policy-level "consensus" to point to when they delete quality images from Wikipedia. In this discussion, I've seen people say Wikipedians "turned a blind eye" to images with regard to NOR in the past, but that phrasing is a deliberately opaque way to say "many years of consensus determined this was fine and I don't like it." lethargilistic (talk) 19:06, 29 December 2024 (UTC) [reply ]
  • Additionally, FWIW, I would take the strong position that it is always appropriate for an illustrator to contribute an illustration of a subject just like it is always appropriate to add new text. Whether that contribution will remain is subject to whether it improves the article. I think this framing is particularly apt for articles about people (living or dead) because readers want to know what subjects looked like. It is one of the most basic things an encyclopedia article about a person ought to include, so adding an image of a person where none existed is effectively always an improvement to the article. So the real question is whether it is an acceptable depiction in the context of coverage of the subject by a free encyclopedia. For instance, if it is easy to source an appropriate photo and include it under either fair use or a license, then the photo ought to be preferred; I don't think anyone would disagree with that. But if no such photo currently exists, I don't think there is a coherent reason to deny illustrations their place in Wikipedia. If one Wikipedian really doesn't like a particular illustration by another Wikipedian, then their basic options are to build a consensus for its removal or to source an acceptable photo to replace it. What's wrong with that? There is no call for a categorical ban on contributing this kind of content because this kind of content fits directly within Wikipedia's mission; it should continue. lethargilistic (talk) 19:39, 29 December 2024 (UTC) [reply ]

    It isn't always appropriate for an editor to add new text. If the next text reports on personal observations, it doesn't belong here. Largoplazo (talk) 22:44, 29 December 2024 (UTC) [reply ]

    As I said, it is appropriate to add text. As you said, the contents of that text might not improve the article. I think this discussion has jumped directly to the second part without appreciating the first with regards to images. lethargilistic (talk) 02:31, 30 December 2024 (UTC) [reply ]

  WP:OR bans entire skillsets from contributing text, if you're going to put it that way, insofar as we don't allow people to report their own observations or first-hand findings either. Are you opposed to WP:OR altogether? Largoplazo (talk) 22:38, 29 December 2024 (UTC) [reply ]

  No, I am not opposed to OR as a policy. I am opposed to the idea that user-generated illustrations are inherently OR, which is the vibe I have gotten from this discussion. Every time someone generates text based on a source, they are doing some acceptable level of interpretation to extract facts or rephrase it around copyright law, and I don't think illustrations should be considered so severely differently as to justify a categorical ban. For instance, the Gisele Pelicot portrait is based on non-free photos of her. Once the illustration exists, it is trivial to compare it to non-free images to determine if it is an appropriate likeness, which it is. That's no different than judging contributed text's compliance with fact and copyright by referring to the source. It shouldn't be treated differently just because most Wikipedians contribute via text.

  Additionally, I suspect we mean different things when we say "entire skillsets," although I admit your message is quite short and you might be taking a stronger position. I think you are referring to interpretive skillsets that synthesize new information like, random example, statistical analysis. Excluding those from Wikipedia is current practice and not controversial. Meanwhile, I think the ability to create images is more fundamental than that. It's not (inheretly) synthesizing new information. A portrait of a person (alongside the other examples in this thread) contains verifiable information. It is current practice to allow them to fill the gaps where non-free photos can't. That should continue. Honestly, it should expand. lethargilistic (talk) 02:54, 30 December 2024 (UTC) [reply ]

This discussion has morphed from a prejudice against AI as a tool to create images into wider prejudice against user generated image content. I'm particularly worried about the discussion to try to find wording that achieves:

How do we write one or two sentences (or even a paragraph) that discourage the use of:

1) AI-generated images 2) User-created artwork (drawings, paintings, sketches, cartoons, etc

that have not been previously published in RS

This is absolutely not and never has been a requirement that an actual similar image has to have been previously published. Indeed, copying a previously published image could well be a copyvio. Current policy is:

Original images created by a Wikimedian are not considered original research, so long as they do not illustrate or introduce unpublished ideas or arguments, the core reason behind the "No original research" policy.

The key test is whether the image misleads or contains novel information or claims, not whether the image itself has never been published previously. That's totally wrongheaded. Although our policy notes the difficulty in the project acquiring images (a professional encyclopaedia would commission artists and photographers to generate images) this isn't actually a get-out for images. Our article text is written in WP:OUROWNWORDS and we rely on experienced editors judging whether our own paragraph of wiki text is a fair summary of the source text. The combination of words, the way the topic is introduced to the reader, the user of wiki links and footnotes, all create free content that is entirely unique to our project. Illustrations are the same. Let's not forget please that Wikipedia is a free content project. Text-contributing editors getting snooty about users who generate our images is not a good vibe. I strongly advise ending this discussion. We should remove images if they fail to adequately and faithfully illustrate the topic or introduce ideas or arguments that are unsourcable. What tool was used to create them or whether they were created by a professional or an amateur is irrelevant to NOR. -- Colin°^Talk 10:34, 31 December 2024 (UTC) [reply ]

When I started this discussion the sourcing for the creation of an image *and* how well, how much that image skews to the cited sources...all of that was and is my issue. Sure we craft cited sources into our own wording but we also adhere to reliable sources in doing so. What should Wikipedia do for placed images - created by AI or not - when the cited source or sources are vague? In one case multiple flags were created, that were based on descriptions in historical texts (mostly newspapers if I remember correctly) but for an example, the description in one case was something like "black letter F in the middle of a white flag". So... which font? How big was the letter? How big was the white background? Many different versions of this flag could be created and all of them wouldn't necessarily be wrong but all of them aren't necessarily right either. It seems to me that Wikipedia might need a policy or guideline or something that when editors come across a similar situation it's not just people being some kind of "snooty text-contributing editor" (thanks so much for that) but fellow editors simply trying to keep Wikipedia's entire content - images and *all* - encyclopedic. - Shearonink (talk) 19:52, 7 January 2025 (UTC) [reply ]

This is a good point. IMO, even if there is no one accepted interpretation of something abstract like "Black letter F in the middle of a white flag," I think it would be encyclopedic to include an illustrated interpretation. However, making decisions about how to render it would involve synthesis. I don't think that would be SYNTH tho, because it's not inventing a connection to be published originally on Wikipedia. I think the image in that case would be best described as "demonstrative," and that there should be some kind of well-placed disclosure in the article that it is one example of what it might have looked like rather than an attempt at rendering what it really looked like. The answer to a sensitive situation like that should not be "no images at all." Additionally, because of the ubiquity of cameras, this sort of edge case would be unlikely to occur with a modern concept. Even people who support an, IMHO, ridiculous photo-only standard must allow for the fact that there were no photos before a certain point; we have many articles illustrated with pictures that may or may not actually represent the subject. Heck, the lead image of William Shakespeare is of someone we're pretty sure is not actually that guy. lethargilistic (talk) 20:06, 7 January 2025 (UTC) [reply ]

I think this whole flag thing needs a dose of historical reality. Or heraldic reality, anyway. There are conventions in the field that make it possible to determine from a textual description what the flag is supposed to look like. For example, with something like "Black letter F in the middle of a white flag", there's a general convention for how large a central device is supposed to be (it fills most of the vertical space, but not all of it). The "font" would have been whatever was typical for that time/place. Yes, you might have to look this up, but no, it's not impossible to get that much right.

Also, until the last century or two, each flag was hand-sewn or hand-painted and unique. Having somewhat different versions of what's recognizably the same design was not considered "wrong". If your "Black letter F" was slightly bigger or smaller than the "Black letter F" on the flag for the next ship/company/building, it didn't matter. WhatamIdoing (talk) 03:15, 8 January 2025 (UTC) [reply ]

That being the case, perhaps the best way to show it is demonstrative would be to declare that and present two or three options. That way, no version of the flag would ever be presented as definitive. lethargilistic (talk) 09:34, 8 January 2025 (UTC) [reply ]

Well said. MichaelMaggs (talk) 17:43, 31 December 2024 (UTC) [reply ]

Agreed. lethargilistic (talk) 18:34, 31 December 2024 (UTC) [reply ]

This puts it much better than I could. Thryduulf (talk) 20:24, 31 December 2024 (UTC) [reply ]

• Per WP:NOTLAW, the written rules themselves do not set accepted practice. Rather, they document already-existing community consensus regarding what should be accepted. So, we should gather examples of current best practice to establish what this is. The main page is a good place to find these as content there gets special scrutiny.

Today, there's an example of this sort (right). This seems uncontroversial and not a significant problem, right?

Andrew🐉(talk) 19:24, 2 January 2025 (UTC) [reply ]

That image is clearly-based on reliable sources and accepted/common knowledge. - Shearonink (talk) 19:52, 7 January 2025 (UTC) [reply ]

No, it's not clear because the image doesn't explain how it was produced or cite any sources. It doesn't seem to have been hand-drawn and so I suppose that software of some sort was used. Andrew🐉(talk) 08:25, 8 January 2025 (UTC) [reply ]

Does File:Dorneywood.jpg cite any sources? How about File:Blank A4 paper.jpg? Why did you upload those if you thought that images need to explain how they were produced and cite sources? WhatamIdoing (talk) 20:10, 15 January 2025 (UTC) [reply ]

The Dorneywood file (pictured) contains metadata which includes the precise time and location. If you put those coordinates (51°33'16.8"N 0°38'53.2"W) into Google Maps you get a nicely detailed plan of the estate showing where I was when I took the photo. I might have added an extensive narrative about my visit there but, normally, no-one would be interested or read it. As it was essentially a point-and-shoot snapshot, there didn't seem to be any need to say more.

The planetary diagram is quite different as it is a novel creation with a mix of fact and fiction shown in quite an artificial and abstract way -- none of it is to scale or with any particular time specified. The choice of objects shown or not shown seems quite arbitrary and synthetic and gives the impression that these bodies would normally appear in fiction together when they are more usually distinct. And the nature of its construction is not clear or obvious.

Andrew🐉(talk) 09:17, 16 January 2025 (UTC) [reply ]

WhatamIdoing, Shearonink, Andrew. Regarding the Fictional planets of the Solar System — the body paragraphs have citations (references) and these descriptions seem to verify that the image is legitimate. Also, four out of the five body paragraphs link to main articles regarding each segment. So, to me, this is acceptable. Regarding Donneywood - the images in that article might need to have some sort of verification. How do we know these are images of the actual house or estate? ---Steve Quinn (talk) 03:32, 16 January 2025 (UTC) [reply ]

The usual rule is that editors attempt to see whether a photo such as File:Dorneywood.jpg looks like what it's supposed to, e.g., by searching for images online. Of course, if there are two identical houses in existence, one might have a photo of the wrong one, which means it would only illustrate what the subject looks like without being indisputable proof that the article's subject exists, but the purpose of an image is illustration, not proof, so I don't think that's a terrible outcome.

The salient principle for me is a version of Hoyle's Law: Whatever the game, whatever the rules, the rules are the same for both sides. So if it's okay for you to upload an image like File:Dorneywood.jpg, and you expect us to trust that you correctly described the contents, then you should extend that same general expectation of competence and truthfulness to the rest of us. WhatamIdoing (talk) 03:59, 16 January 2025 (UTC) [reply ]

I don't extend unqualified trust to images but judge each case on its merits. And I don't mind if people challenge my own images. For example, I took a picture of some seat fabric (pictured) which sparked an interesting discussion with a keen-eyed fanatic. They spotted that the date couldn't be right and so it proved -- the camera's clock was a day off. Andrew🐉(talk) 09:55, 16 January 2025 (UTC) [reply ]

OK, so they judged the photo by comparing it to facts outside the photo. In that case, probably their first-hand knowledge. In the more typical case, it would be non-free photos. That's just a common-sense application of verification in the image context. Nobody is asking for unqualified trust. Personally, all I'm advocating for is to treat images no worse than we treat text. lethargilistic (talk) 10:27, 16 January 2025 (UTC) [reply ]

I absolutely agree we need an explicit policy banning user-generated solely-text-based illustrations and am baffled why this would be controversial. Illustrations that are based on other reliably-published illustrations are clearly distinct from those based on only interpreting text/unpublished images; the latter should be prohibited for the same reason we already prohibit textual material sourced from the editor or from non-expert SPS. I would also argue that if no professional has been interested enough to publish their graphical interpretation of a text description yet, then a graphic isn't BALASP in the first place. JoelleJay (talk) 23:28, 7 January 2025 (UTC) [reply ]

JolleJay, Here! Here! Well said. The key here is, already "reliably-published." The other stuff that is essentially original research from interpretation probably should not be allowed. ---Steve Quinn (talk) 06:54, 8 January 2025 (UTC) [reply ]

I'm surprised this discussion is ongoing. The AI is a tool, operated by a human, who is responsible for the results. Sometimes the results can be embarrassingly awful and should never have been uploaded and inserted. That true for AI or hand illustrations or for operating a camera or a scanner.

If the results are misleading or advance ideas or concepts or make claims unsupported by reliable sources, we have policy to remove the image from an article. If not, then we simply don't care how the image was made.

The moment our AI overlords start uploading images and inserting them into articles all by themselves, then we can start having rules to ban AI. I don't think we are there yet. Colin°^Talk 08:58, 16 January 2025 (UTC) [reply ]

Here's another example from today's main page. It looks simple but the image details explain that This image is a focus stacked image consisting of 23 images that were merged using software. As a result, this image underwent digital manipulation which may have included blending, blurring, cloning, and colour and perspective adjustments. As a result of these adjustments, the image content may be slightly different than reality at the points where the images were combined. This manipulation is often required due to lens, perspective, and parallax distortions.

The key point is that it may be "different than reality" but it's a featured picture.

Andrew🐉(talk) 08:30, 8 January 2025 (UTC) [reply ]

We have no way of knowing whether the merging was or was not carried out with AI assistance. For some people it seems that an image that is identical, down to the pixel, would either be perfectly acceptable if not done using AI but cause the sky to fall in if AI was involved at all. It's utterly ridiculous. Thryduulf (talk) 11:42, 8 January 2025 (UTC) [reply ]

Not sure where I should place this discussion, but I hope I'm at the right place. It is often said that interviews are "primary" sources, meaning they are not reliable per WP:PRIMARY. However, most of the times we get personal information (birth dates, birth place and backstory) and upcoming release dates for movies and music from interviews (late-night shows and so on) and they always turn out to be accurate. I think if the interview was published by a reliable source then it's most definitely reliable, because if another publication quotes that interview, no one would say it's not reliable. Not sure if I make much sense, but any objections? dxneo (talk) 23:10, 26 December 2024 (UTC) [reply ]

Re "they always turn out to be accurate": [citation needed]. Sources directly from the subject of a biography, such as in interviews, can be reliable for uncontroversial factual claims such as birthdates per WP:BLPSELFPUB, but should not be used for evaluative claims nor when there is good reason for skepticism regarding the claims. Even for birthdates, people can often falsify these out of vanity or out of pressure from whatever industry they're in to be a different age. —David Eppstein (talk) 23:40, 26 December 2024 (UTC) [reply ]

@Dxneo, Primary is not another way to spell 'bad'. You should avoid trying to build an entire article exclusively on primary sources (though this is pretty common for discographies), but you may use reliable primary sources to fill in ordinary or expected details. If you are at all uncertain about the material, consider using WP:INTEXT attribution: "In an interview with Music Magazine, the musician said she was born in California" or "According to Joe Film, the movie will be released in September 2025". WhatamIdoing (talk) 05:51, 27 December 2024 (UTC) [reply ]

I understand why we have an exception for images in this policy - we have a limited selection of free images, so we need to rely on user-uploaded content. So if someone uploads a photo they took of a celebrity, that is fine to include in the article since it's not considered original research.

But what happens if someone claims their image is of a certain celebrity but other editors dispute it? It seems like we have limited recourse within policy to handle that. The image doesn't really illustrate or introduce unpublished ideas or arguments, it is just an image of a person that has possibly been mislabelled. Would it make sense to revise the first paragraph of WP:OI to the following? The last sentence is new:

Because of copyright laws in several countries, there may be relatively few images available for use on Wikipedia. Editors are therefore encouraged to upload their own images, releasing them under appropriate Creative Commons licenses or other free licenses. Original images created by a Wikimedian are not considered original research, so long as they do not illustrate or introduce unpublished ideas or arguments, the core reason behind the "No original research" policy. Image captions are subject to this policy no less than statements in the body of the article. Additionally, images whose authenticity is disputed may be removed in accordance with consensus.

– Anne drew 16:42, 27 December 2024 (UTC) [reply ]

Good point and good start. But "removed in accordance with wp:consensus" is unclear. Do you need a consensus to remove? Do you need a consensus to keep? North8000 (talk) 20:10, 27 December 2024 (UTC) [reply ]

In the case of a living person, we need to take extra care to "get it right"... therefore we would default to needing a "consensus to keep" if there were a disagreement over the image. Blueboar (talk) 20:20, 27 December 2024 (UTC) [reply ]

I don't we need to write it out on the page, WP:ONUS is pretty clear. Traumnovelle (talk) 21:05, 27 December 2024 (UTC) [reply ]

I don't think the addition is necessary: In all cases, for all images, for all material, for any reason, consensus can force removal.

One of my touchstones for this policy is a dispute years ago with a since-blocked AIDS denialist. Look through Wikipedia talk:Verifiability/Archive 40#RfC: Do images need to be verifiable? for one of the discussions. AFAICT he wanted certain images removed from Wikipedia because the existence of a photomicrograph of a virus undercut his story that these viruses don't exist, but since that's not a policy-based reason, he generally asked for images to be removed if there was no source to authenticate the contents. We had "images whose authenticity were disputed" – but only by a POV pusher. I would not wish to give him, or POV pushers like him, a rule that says that disputing authenticity is his best path to removing the image. I think we could safely predict that this addition would turn into a WP:BEANSY recommendation to partisan editors to dispute the authenticity of all unflattering photos of their favorite politicians.

What I'd suggest instead, in these cases, is relying on WP:PERTINENCE, which says "Images should look like what they are meant to illustrate, whether or not they are provably authentic." In the situation described above, we have an image of a BLP that editors dispute. Why do they dispute it? I'd guess it's because it doesn't look like the person. I'd bet that most of us have had the experience of a photo not turning out the way we expect, and even though we know with absolute certainty who is pictured in the photo, we couldn't say that the photo is representative of the person. That might be the only thing that's going on in this photo: Right person, but odd angle, odd expression, odd lighting – and the result is that the image doesn't look like what it's mean to illustrate, and therefore should be rejected per MOS:IMAGES. One doesn't even have to dispute the authenticity to do this: just say that it doesn't look like what you/readers expect, and the guideline therefore rejects it. WhatamIdoing (talk) 22:12, 27 December 2024 (UTC) [reply ]

Thanks for the detailed reply! I hear you on using WP:PERTINENCE in these cases. That's what I leaned on in a recent discussion around a disputed photo of a tornado. It just seemed like a bit of a workaround, having to first dispute the verifiability of the caption, and then separately the pertinence of the image itself. I think expecting editors to formulate an argument like that using multiple policies/guidelines is asking a lot.

But maybe I'm overthinking this. To your point, it's already the case that consensus can remove disputed photos. But I do think there would be some value in explicitly stating that WP:OI isn't intended to help retain inauthentic photographs. For what it's worth, this isn't a one-off issue. In a similar discussion this month, I incorrectly relied on WP:OR to support the removal an image with disputed authenticity. – Anne drew 22:30, 27 December 2024 (UTC) [reply ]

The question with the tornado isn't whether we're "helping retain inauthentic photographs"; we'll make a decision by consensus.

I wonder, though, why your response was to remove a probably-but-not-definitely authentic photo, instead of placing it in proper context? For example, one compromise approach – neither unquestioning acceptance nor removal – would be to remove it from the infobox and add a caption that says something like "Very few images of this storm exist; this photo has been claimed on Twitter to be authentic". WhatamIdoing (talk) 04:42, 28 December 2024 (UTC) [reply ]

Hmm you raise a good point, just changing the caption would be enough to avoid misleading readers. I'll update my !vote accordingly. Thanks – Anne drew 17:18, 28 December 2024 (UTC) [reply ]

One issue here (as I said in that discussion) is that it's a photo from social media uploaded under fair use with the only sources attesting to its identity being comments on Twitter and Reddit. People on social media have been known to misattribute images of tornadoes before. An I think saying "this is probably the tornado but we're not sure" undercuts the article. TornadoLGS (talk) 21:39, 28 December 2024 (UTC) [reply ]

Yes. Subjects that require expertise to identify or accurately characterize should not be treated the same as subjects that are easily verifiable (like a picture of a Walmart, or of a human hand). If a photo isn't even an editor's own work, and the only attestations to its identity are random social media comments, including it in an article where it will be scraped by Google and placed at the top of image search results is a disservice to more than just our readers. JoelleJay (talk) 23:43, 7 January 2025 (UTC) [reply ]

Claims of own work might not be that reliable either. I've been pondering File:Louis-Charles Damais.jpg for the past couple of days. It has a strange contrast, but it doesn't not look like the subject, if you know what I mean. The uploader added this image, edited the relevant fr.wiki article, then vanished. The metadata says it was modified with photoshop in 2021. One thing I am reasonably sure of though is that it's unlikely that this photo of someone who died in 1966 is the own work of an uploader in 2022. CMD (talk) 02:03, 8 January 2025 (UTC) [reply ]

 You are invited to join the discussion at Wikipedia:Village pump (policy) § BLPs. Some1 (talk) 00:17, 1 January 2025 (UTC) [reply ]

• Pages with missing files