Commons:Deletion requests/File:Scene of ancient cave humans.jpg

This deletion discussion is now closed. Please do not make any edits to this archive. You can read the deletion policy or ask a question at the Village pump. If the circumstances surrounding this file have changed in a notable manner, you may re-nominate this file or ask for it to be undeleted.

File:Scene of ancient cave humans.jpg

Inaccurate anatomy and anachronisms, AI simply can't do this kind of stuff which needs critical thinking, Commons:out of scope and misleading. FunkMonk (talk) 09:42, 8 April 2024 (UTC)

 Keep I understand frustration with the prevalent thinking AI could do anything and things that need critical thinking; however this image is not meant to show "This is how ancient humans may have looked in the past" but "This is a demo illustration of how AI could in the future be used to create high-resolution art of ancient humans".

Maybe the description could be clearer. I don't see inaccurate anatomy and a new version of the image could be created that solves these problems (for example the description includes I will try to fix the issues with the faces and so on; there probably already is AI for that and I think I already know something that could work. If you can improve the image, please do.). There's lots of media Category:Inaccurate paleoart. The image is just meant to be a AI-type-upscaling of File:Paleolithic period.png where the critical thinking is done by the creator of the base-image. Another idea is to use AI for paleoart by a) feeding it with lots of images b) then selecting a good result c) then modifying the image using AI tools as well as image editors to correct any issues to produce a good-quality image without issues. This is basically the only illustration that could be used for the concept, of good enough quality, not out of scope, and useful. The linked png image is of such low-resolution that it's hard to set your mind to imagine yourself living daily life in that time period for which a high-resolution image would be useful but is otherwise missing in WMC where this image can also be used to highlight that and explain one way this gap could be closed. Inaccuracies and issues can and are warned about in the file description which could be improved. Prototyperspective (talk) 10:41, 10 April 2024 (UTC)

 Keep Anthere (talk) 15:10, 11 April 2024 (UTC)

 Delete – Personal artwork with no academic value. There is no point in generating paleontological images using AI. It will always be inaccurate. Ta-tea-two-te-to (talk) 01:21, 5 May 2024 (UTC)

You replied at all the deletion discussions with the same explanation as SlvrHwk just a few hours before. What's your association with that person, did you send the person a message about the DR or something? It seems suspicious. And for your explanation, it's quite simply false because it can be modified in an image editor or be corrected otherwise so it's not always inaccurate as Sora's Woolly Mammoth video already proved. Prototyperspective (talk) 10:15, 5 May 2024 (UTC)

I never said Sora's mammoth was accurate. Since this is not a human-drawn drawing based on actual anatomical features, it is inaccurate. Ta-tea-two-te-to (talk) 21:13, 5 May 2024 (UTC)

Since it is not a hand-picked berry, it can't be food; since it's not a hand-woven clothing, it can't fit a human. This is the kind of void reasoning you provide? Moreover, I mentioned human-modified AI images. Prototyperspective (talk) 14:11, 6 May 2024 (UTC)

Regardless of whether or not these are human- or AI-made, they and other user-made reconstructions will need to pass by the paleoart or dinoart review. The Morrison Man (talk) 11:50, 7 May 2024 (UTC)

 Delete – Per the above; this image is full of countless issues that make it entirely unusable. Suggesting that AI can be used to upscale low-resolution images to make a more immersive experience entirely misses the point. High-resolution inaccuracies are still inaccuracies and do nothing to enhance our understanding of prehistoric ecosystems or their inhabitants. This push to use artificial intelligence to create paleoart is full of fundamental flaws and misconceptions. While I recognize that AI has its uses, attempting to apply it as an autonomous tool for scientific reconstruction is misguided and misleading. Accuracy in scientific illustration requires more than just algorithms; it demands the irreplaceable insight of those with genuine scientific and artistic experience. Arguing that you can simply feed AI images, select the best output, and then refine it to correct any inaccuracies, is flawed. Usable paleoart requires a reasonable understanding of anatomy, physiology, and biomechanical constraints. The claim that giving AI training data will suffice to bridge this knowledge gap is insubstantial and it misleads people regarding how scientific inference is conducted. Saying that new versions of these images can be tweaked to 'fix' inaccuracies related to anatomical features implies an approach where egregious errors in basic anatomy can simply be ironed out by the very tools that introduced them in the first place. Each error in an AI image has the potential to propagate misunderstandings of the evolution, form, and function of the depicted organisms. Finally, uploading AI images to Wikimedia, where they are readily accessible and usable by the general public, is concerning. Not only does it spread misinformation, but it also crosses the ethical boundaries of scientific communication. There is no need for poorly made AI images simply to illustrate the fact that AI cannot make them. The availability of these images— flagged with caveats or not—spreads a distorted view of the past and ultimately undermines the educational and academic value of paleoart. SlvrHwk (talk) 08:15, 5 May 2024 (UTC)

Your claims are based on the premise that these are autonomous tools which is deep misunderstanding. These are tools, software that a human person uses. And these errors can be fixed with AI tools or simply an image editor. Sora's Woolly Mammoth video is proof that it can be sufficiently accurate. I see how you're offended but useful nearly unique images shouldn't be deleted because people misunderstand the tools and intended usefulness and are offended by the novel toolsets used to create them. The inaccuracies in the image are not misleading where it looks realistic but is actually false, rather it's objects you can't discern as any objects and so on. People who use these tools can have "a reasonable understanding of anatomy, physiology, and biomechanical constraints" and the resulting images would greatly benefit from that, this is just an illustration conceptually. It's a unique illustration how AI tools could be used by people for useful images. In any case, COM:INUSE applies. Prototyperspective (talk) 10:11, 5 May 2024 (UTC)

 Delete - As previously stated, this image has no scientific value and its existence on the site is detrimental to the perception of prehistoric humans. It's very telling to see the excuse that "The image is just meant to be a AI-type-upscaling of File:Paleolithic period.png where the critical thinking is done by the creator of the base-image.". That statement ignores the fact that the "base image" is a century-old painting influenced by the directives of a notorious white supremacist (more information in this thread: [1]). This is a great illustration of the problems with this kind of AI "remix": every piece of paleoart has a cultural context and a gap between the artist's intent and available knowledge. The gap in this case just happens to be basic human decency plus about 100 years of paleoanthropology research.

It's also well understood that, regardless of the level of detail present in an AI prompt, major discrepancies and artifacts will always appear in the resulting image. Extra toes and fingers are just one obvious and frequent example. The Sora video stumbled into something recognizable as a woolly mammoth only because it picked an animal which already has a huge sample of pre-existing public imagery which mostly conforms to scientific evidence. That has everything to do with the human-made imagery already collected by Sora, and much less to do with the details of the prompt. No amount of prompter experience or adjustment will result in images of the same quality as actual (i.e. human-made) paleoart. 99.999% of extinct animals don't have enough human-made imagery for an AI crawler to properly interpret, even something as famous as a dodo ends up with extra toes. We'd be better off giving the same prompt to a kindergartner and accepting whatever they produce. AI does not make paleoart "more accessible and less resource-intensive to produce", it just allows low-quality prehistoric animal imagery to be generated at a faster rate. A real paleoartist will always have a deeper understanding of the subject matter than even the most comprehensive prompts and training data. There's a basic cost-benefit analysis for why paleoartists don't use AI: the time spent editing or "refining" an AI prompt (pre-generation) and/or image (post-generation) is wasteful for someone who can personally create something of a higher quality on their own in the same amount of time. And that's not even touching on the ethical problems with AI tools. COM:INUSE does not apply, the only public page which uses this file is a tutorial of Prototyperspective's creation which advocates for widespread adoption of AI imagery within Wikimedia Commons. There's an obvious conflict of interest here. NGPezz (talk) 03:17, 6 May 2024 (UTC)

Again, why do you assume it must have "scientific" value? Lots of media here does not have scientific value. You and the other delete votes are largely based on such flawed assumptions.

That statement doesn't ignore that fact/… It's an illustration how people could create an accurate low-res painting and scale it up to high-quality using AI. Since I couldn't find a better image, I used this one, which is widely used across Wikipedia. Please answer where exactly I claimed that the person who did the base image did a good job at critical thinking / the image?

It's also well understood that these things can be easily fixed, for example using an image editor or negative prompts which now commonly include phrases like "extra fingers".

Please point to where I exactly I claimed that it has more to do with the prompt? I even explained that the used training data is very important and the resulting quality depends on that.

In short your explanations is as much based on misunderstandings, flawed assumptions, and falsehoods as the prior comments that apparently by coincidence just appeared very shortly before yours.

COM:INUSE does apply. Even if you want to selectively ignore my wikibook as a legitimate use, which it is and it doesn't "advocates for widespread adoption of AI imagery", it's also in use on wiktionary. The conflict of interest is with three users making very similar comments shortly after another, all of which uploading paleoart and offended by people illustrating a novel tool for it that they don't use and object to. Prototyperspective (talk) 09:30, 6 May 2024 (UTC)

Paleoart is an educational medium, using AI to create it is like using AI to write a scientific paper or recreate a historical photograph. A high level of scrutiny is necessary. Neither the base image nor the AI "upscaling" are accurate enough to be educational, and they don't actually resemble each other anyways. Calling the AI version an "upscaling" is misleading, and like I say, that's before even considering the ethical problems of the tool. There are actual cases where someone tried to upscale a low-res paleoart snippet into a high-res product with AI, and the results were disastrous (the older version of the file here). Neither the intended role of this technology nor its current state are satisfactory for educational purposes. All of your arguments boil down to the belief that this tool is not fundamentally flawed, but simply not used to its maximum potential, requiring better base images or more comprehensive prompts or extensive post-generation editing in order to properly compete with traditional methods it hopes to usurp. That belief (emphasis on belief) has no merit on Wikimedia Commons or the wider world. Outside of your personal pet project, the image is in use on exactly one wikidictionary page in exactly one language, it was probably put there entirely by mistake by someone who doesn't understand the ethical or educational problems with AI art.NGPezz (talk) 16:03, 6 May 2024 (UTC)

The link is interesting as are those elaborations.

No, the tool may possibly be fundamentally flawed (and I already do think text-generators are fundamentally flawed) but it can still be used to produce an image that gets reasonably accurate and can then be edited if selecting the most accurate image from a larger set of images as well as modifying it with further AI prompts doesn't suffice. Thanks for the explanations, I still maintain it should be kept at least as long as there's no better image because having it here doesn't mean it has to be used or when used can't be annotated in a way that makes it clear it's very inaccurate. If there was better images, I'd also replace the images in the wikibook; and the text there I think already says it currently can't do so well but this could be a possible application (image serves to illustrate the idea). There's lots of media and info about concepts that may work in the future but don't work now or the time depicted. "Ethical problems" which are ethical problems in your personal subjective opinion (I think it would be unethical to not make use of ways to save time in a world of major problems to solve and available time-intensive training data) – it's not a reason why images here are used or not. And the use in the wikidictionary page is legitimate. Your objections to AI are not a valid reason to delete useful media or as valid as deleting all photos because 'photographs kill paintings' (so these are bad and we should delete them). Prototyperspective (talk) 10:07, 7 May 2024 (UTC)

 Delete Gen AI is fundamentally incompatible with paleoart because gen AI has no knowledge that it can apply to make correct and informed reconstructions.

Using gen AI for the purpose of paleoart is misinformation. Plain and simple. There is no fixing it. SuspiciousHadrosaur (talk) 12:44, 7 May 2024 (UTC)

Again, the user of these tools can select an accurate image and/or modify it so that it's accurate. The use of AI here is part of the process not the whole process. Prototyperspective (talk) 13:42, 7 May 2024 (UTC)

 Delete "However this image is not meant to show "This is how ancient humans may have looked in the past" but "This is a demo illustration of how AI could in the future be used to create high-resolution art of ancient humans"."

This image therefore is for your own personal use, and not intended to have any educational or scientific value whatsoever. AI images, especially used for paleoart and on a website regarded as a research tool, are harmful, and uploading it here could affect people looking for information on this topic as it is inaccurate to the subject at hand. HolesGuy (talk) 15:27, 7 May 2024 (UTC)

You name a specific (instead of unspecific and doubtful as the case for countless other DRs) educational purpose to make a point it doesn't have one? An illustration doesn't have to be perfect to be useful in principle. Also you didn't provide any reason before your "therefore". I would see how could affect people looking for information on this topic would make a good point for why e.g. the file-title should be renamed to sth like "AI-made low-quality …" instead of just having this info in the file description. Prototyperspective (talk) 16:43, 7 May 2024 (UTC)

A minor note for clarity: delete-voters seem to assume art showing ancient scenes can only have scientific accuracy as a purpose. But the core concept for which AI-application has been illustrated is not some kind of extremely detailed extremely accurate image but a high-resolution image that shows a few key concepts (e.g. living at a cave) in an immersive way that can help the viewer to imagine the past.

If some documentary or even fiction film about ancient Roman cities shows buildings with some architectural features that didn't exist at the time, it doesn't diminish the value of the scene compared to blackscreen even if some archaeology experts may be greatly upset by the inaccuracies.

It's not the point of it, it can be pointed out, and I think most (if not all) of the issues here are not misleading or false but things like misgenerated arms or blurred faces (clearly inaccurate; these could be fixed in a later version btw). Prototyperspective (talk) 14:30, 8 May 2024 (UTC)

I'd say that your documentary comparison makes no sense here. We'd expect a documentary about ancient Roman cities to be accurate, same as we expect paleoart to be accurate. As for fiction films, the focal point in those is most often not on the backgrounds but on the characters in them, so that also falls flat. The Morrison Man (talk) 16:29, 8 May 2024 (UTC)

First off I just added this for clarity, things relating to the particular delete vote are mostly in the prior reply. It does make sense because I was talking about the level of (in)accuracy so you missed the point. Maybe don't call it paleoart then especially since this is not even in a paleo-illustrations cat but one about "art". I really don't know why my points are so hard to get across so here's an example: in this scene of a documentary about a day in ancient Rome there may be several (relatively) minor inaccuracies such as unlikely or impossible clothing and building features or glitches like the blurred faces here. It's still valuable and helps the viewer to consume the other educational content as well as imagine and visualize daily life at the time. And I'd watch such fiction films for imagining the past not any characters set into that timeframe, the dismissal doesn't make sense and nothing of it falls flat. Prototyperspective (talk) 16:59, 8 May 2024 (UTC)

---

Deleted: per nomination & informed discussion. --Infrogmation of New Orleans (talk) 20:49, 8 May 2024 (UTC)