User talk:Adamant1

---

Commons:Deletion requests/Files in Category:AI-generated furry

I kindly request that you add your AI deletion requests to the appropriate category next time. Otherwise, it becomes harder for people to find Trade (talk) 23:51, 19 December 2023 (UTC)

@Trade: I usually add categories, but I wasn't sure which ones would be appropriate in this case since I don't usually work in the area. Although I see now that there's a category specifically for AI artwork. So I'll be sure to add it next time. Thanks. --Adamant1 (talk) 23:55, 19 December 2023 (UTC)

Seasonal Greetings!

Merry Christmas and a Prosperous 2024!

Hello Adamant1, may you be surrounded by peace, success and happiness on this seasonal occasion. Spread the WikiLove by wishing another user a Merry Christmas and a Happy New Year, whether it be someone you have had disagreements with in the past, a good friend, or just some random person. Sending you heartfelt and warm greetings for Christmas and New Year 2024. Happy editing, A1Cafel (talk) 03:39, 25 December 2023 (UTC) Spread the love by adding {{subst:Seasonal Greetings}} to other user talk pages.

A1Cafel (talk) 03:39, 25 December 2023 (UTC)

@A1Cafel: Thanks. You to. --Adamant1 (talk) 07:35, 25 December 2023 (UTC)

Flickr2Commons

Saberia informar o problema que houve com o Flickr2? Estou tentando baixar imagens, mas sem êxito. Luiz79 (talk) 00:32, 1 January 2024 (UTC)

@Luiz79: I don't use it. So I can't really say if it's working or not either way. You might ask about it on the Village pump or Commons talk:Flickr2Commons though. Sorry. --Adamant1 (talk) 02:10, 1 January 2024 (UTC)

Antonio Carbonati

Hi, I replied to you about Commons:Deletion requests/Files in Category:Etchings by Antonio Carbonati. Please tag me in your answers to see the notification Moxmarco (talk) 13:04, 4 January 2024 (UTC)

Nice uploads

I just wanted to say that I like all of your postcard uploads. Thank you for contributing them and taking the time to share them.--SDudley (talk) 01:43, 9 January 2024 (UTC)

Voting twice

I don’t know if you’ve noticed but you voted twice on Commons:Deletion requests/Files uploaded by Hyju. Don’t do this again. Dronebogus (talk) 18:15, 16 January 2024 (UTC)

I didn't notice. That can happen when the conversation is just a wall of text by one user. --Adamant1 (talk) 18:20, 16 January 2024 (UTC)

A “please” might have been nice. I’m sure it wasn’t done intentionally. - Chris.sherlock2 (talk) 08:42, 24 January 2024 (UTC)

Ebay coin/ token photo licensed as PD (Public Domain)

Your upload here: File:Brown's Cigars & Tobacco token front.jpg

I looked at the link you provided, but I do not see any license. Could you provide a link to Ebay's policy related to individual account owner's posting their own photos or scans? Thanks, -- Ooligan (talk) 18:31, 16 January 2024 (UTC)

@Ooligan: The coin was created in 1910 and the photograph is to generic itself to be copyrighted. Usually in those cases we just assume PD-US-expired is appropriate. Like for example the many scans of postcards that are copied from places like eBay and Flickr even though the person who did the scan might (usually wrongly) think their version is copyrighted. Although if you want to ask for clarification about it on the Village pump be my guest. I'm not going to lose sleep over it if the rules end up being different for coins then postcards or photographs of other items where that seems to be how we do things. Thanks. --Adamant1 (talk) 18:45, 16 January 2024 (UTC)

Adamant1, thanks for the explanation. It seems like there is great potential to upload many more quality photos from Ebay. Again, thank you. -- Ooligan (talk) 20:29, 16 January 2024 (UTC)

Obscenity debate

FWIW, I reread your comments and you actually did cite your source properly. I apologise, I missed the bit where you did so. Overall, a well reasoned argument… with a source! Definitely not bullshitting. - Chris.sherlock2 (talk) 08:40, 24 January 2024 (UTC)

AI art

My understanding is that generative art uses an existing corpus of images to generate the images. If this is the case, then if any non-free images were used to train the dataset, would that potentially make the artwork a derivative artwork? In that case, I’m curious how Commins could accept the artwork… I mean, we don’t allow fair use images for similar reasons, right?

It follows if this is true that if we don’t know the service that generated the image, then we can’t know its dataset (and similarly if we do know the service and don’t know the dataset) and thus we can’t determine if the images used to train the service were all free! And if we do know the service and we do know the dataset, and it includes non-free images then we would not be able to use any images generated from the service.

Does this make sense? Is this a reasonable conclusion? Thought I’d get your take as you seem to have a good handle on this matter. - Chris.sherlock2 (talk) 11:59, 24 January 2024 (UTC)

(talk page stalker) It's more complicated than that. It's presumably analogous to the issue of a human artist creating an "imaginary" image of a real person, building, etc. Here are a few cases:

• Copyrighted monument or building in a country with no freedom of panorama (e.g. the Memorial of Rebirth in Bucharest, or the Fondation Louis Vuitton in Paris): there is presumably no way to create a "free" image of this. Any image that identifiably shows this monument/building is going to violate the sculptor/architect's copyright.
• A building that is itself either in the public domain or where FoP allows images (e.g. the White House in Washington, D.C. or the Modern Art Museum of Fort Worth: if a human were to be drawing this, they could, in principal, either (a) draw from life, (b) draw from one copyrighted image, (c) draw from one uncopyrighted or free-licensed image, or (d) draw from a variety of images, some of them copyrighted. Clearly case (a) and (assuming any licenses are complied with) case (c) are fine, and equally clearly case (b) is not. Case (d) is the tricky one, and it is exactly analogous to what will typically happen with generative AI unless a prompt steers it toward working from some particular source. The question is always going to be: was it unduly influenced by one unacknowledged source (or a small number of unacknowledged sources). Note that without some criterion like that, no one could draw (for example) a picture of the White House because we've all seen (and presumably been influenced by) numerous copyrighted images of the White House.2

There are other cases, but I think that shows why this is complicated. - Jmabel ! talk 20:01, 24 January 2024 (UTC)

@Chris.sherlock2: I pretty much agree with Jmabel's points. An AI generated image of say something like an otherwise copyrighted building in a country without FOP would obviously be copyrighted. I've also seen plenty of instances where the backgrounds in images are clearly cropped from previous photographs, which I assume were in the original training set. I assume they would be copyrighted also.

You can get into some grey with images that are based on original IPs though. It's not really clear to me where the line is with something that is clearly based on exiting characters or stories but might not be a 1/1 recreation of either one. Like with File:Sonic Diffusion art of anthropomorphic cat in rain.png. That image was clearly based on the Sonic cartoon universe and looks a lot like several characters in it. Although I can't point to exactly which one and when I nominated it for deletion @Prototyperspective: argued that something being based on preexisting characters does not make it a problem. So who knows. Although I'd argue the risk to reward makes it not worth keeping images like that one, because really, what's being gained eductionally by keeping that image on Commons? Absolutely nothing.

There's also other things to consider besides purely copyright. For instance the inherent lack of a source with anything generated by AI. Although certain people would argue that AI generated artwork is no different then something created in Photoshop. The source of an image is clearly different then what software it was created with. We certainly can't say the source of a photograph is Photoshop. So I don't why it would be acceptable to cite an AI generator as the source of AI generated image. At least with someone uploading a drawing they made in Photoshop you can ask what they based their artwork on. In the case of AI artwork there's no way to do that since the training data isn't available to the public.

I think people wrongly conflate a human work being inspired by past experiences, with how AI art generators create duratives by combining multiple previous works into an original. The word "original" doing a lot of heavy lifting of course. Although I don't think the risk is worth the benefit in most cases anyway. As a lot of the AI generated images people upload to here aren't even worth the processing cycles that were used to create them. It's not like we to allow for everything just to host the 1% that are (or might be) educational either. --Adamant1 (talk) 21:29, 24 January 2024 (UTC)

A brief note since I was mentioned; in the DR I agreed that this image you linked here should probably be deleted…I just said that basing it on pre-existing characters is not a problem, otherwise all contents of Category:Middle-earth fan art would need to be deleted since they're based on characters of Tolkien's universe (mainly the books).

The source of an AI image is the user tasking the AI where AI play the major role but both AI prompter / prompt engineer and the software should be named I think. If LAION-5B is all StableDiffusion uses, that dataset is public. The tools learn to 'understand' what is in the image via attached texts so new texts creates new visuals that match them starting from random noise; artists usually don't specify their inspirations and they usually if not always don't know them all anymore either since their learning stretches back to when they where infants where they learned how objects look like and were later influenced by lots of different copyrighted works somewhere in the back of their mind to different degrees in respect to the new artwork they created. Diffusion from random noise is not combining previous works into a new original similar to cut-and-pasting parts of existing images, it's a small-sized AI, no images included, that has learned the concepts in the images and creates it anew. If you deliberately write a prompt that is just the name of a character from a film franchise, it'll likely create a copyvio; but otherwise people can use these tools to create anything imaginable using its lexicon of terms it understands. There are large risks with cutting down on general purpose tools as useful and general as these. Prototyperspective (talk) 22:51, 24 January 2024 (UTC)

@Prototyperspective: Just to clarify, I wasn't saying images based on pre-existing characters is a problem in general, I was saying the specific images that I nominated for deletion are problem because they are based on pre-exiting characters. Images based on pre-existing characters are a problem when they are close enough to the original to be duratives and not when they aren't. Know one claimed otherwise. I certainly haven't and nowhere have I said images based on pre-exiting characters in general should be nominated for deletion or otherwise don't belong on Commons. I think those specific images are close enough to characters in the Sonic franchise to justify deleting them though. I'm sure you get the difference.

Random noise is not combining previous works into a new original similar to cut-and-pasting parts of existing images I can't speak for other AI image generators, but I've generated thousands of images with DALL-E at this point and it clearly crops images from original photographs if you have it generate the same scene enough times or if what your asking it generate is "niche" enough that it was only trained on a small set of images to begin with. Usually it will add other elements to the scene that are clearly AI generated, but that doesn't mean the cropped photographs aren't recreations of the original photographs in the training set or copyrighted. There's nothing inherent to diffusion models that would keep an AI image generator from doing that either. Like if it's only trained on 5 images to begin with then sure it's going to "learn to 'understand' what is in the image via attached texts" or whatever, but the resulting image will still be heavily based on the 5 original photographs regardless. You act like if someone asks an AI generator for an image of a spaceship landing on Mars that it will partially (or largely) be based on images of cats eating kibble cat food. That's not how they work. --Adamant1 (talk) 23:42, 24 January 2024 (UTC)

Adamant1, Prototyperspective and Jmablel thank you for your respectful and informative responses. The issue is definitely more complex than I had first realised! - Chris.sherlock2 (talk) 19:14, 25 January 2024 (UTC)

[cross-posted] On the derivative character drawing thing: usually (maybe always?) no problem if it's based on a verbal description of a character; another matter if it's based on another drawing or other image. So, my own image of Gandalf based solely on Tolkien's text would be fine, but if a Tolkien drawing is still in copyright then anything derived from that would be a problem. - Jmabel ! talk 19:16, 25 January 2024 (UTC)

Adamant1 so... what is the threshold at which the AI generated image no longer becomes a copyright infringement problem? One thing I would have thought would that we need to actually need to see the training data set used to generate the AI image. If we don't know that, then we cannot know if the image is truly free, at least that's what I would have thought. - Chris.sherlock2 (talk) 21:53, 25 January 2024 (UTC)

JFYI you don't decide that even if you thought much about it and think it's an important interesting subject to you. Copyright law decides that and everything so far shows these are not infringing copyright even if a few artists allege that and fail.

And no, it does not cut and paste parts of images which it doesn't have packaged with the software. If you prompt specifically something for which it, roughly speaking, only had seen one image matching the concept, it'll look similar to that one, just like you can use your skills to draw portraits to draw one exactly like a painting that already exists. Prototyperspective (talk) 22:08, 25 January 2024 (UTC)

JFYI you don't decide that even if you thought much about it and think it's an important interesting subject to you. I could say the same about literally every comment you've ever made related to AI artwork. 99% of what you write is pure speculation and conjecture. Even the article you linked to in defense of your personal opinion that the law decided AI artwork can't be copyrighted says nobody knows what the copyright status is or will be. The difference between you and me is that I'm more then to admit that it's all speculation at this point and that there multiple opinions about it out there, all of which have their pros and cons. Whereas all you do is act like your opinion that AI artwork can't be copyrighted no matter what is the correct, god given truth and that anyone who disagrees with you is just ignorant about the topic.

That really goes for your general attitude about this to. Your response to me saying AI generators cut and paste parts of the images that they are trained on sometimes is another example of that. What actual evidence do you have that they don't do that sometimes outside of using just making false analogies to how human's create art? Because I haven't seen any and you'll sit here all day speculating about this and then attack anyone everyone else for doing the same exact thing. So again, what evidence do you have that AI art generators don't cut and paste parts of images that there are trained on when generating images sometimes? --Adamant1 (talk) 22:21, 25 January 2024 (UTC)

I'll be surprised if courts rule that the question of whether Image A infringes on the copyright of Image B is going to be in any way affected by whether Image A was created by generative AI, and astounded if they not only make a distinction but allow more leeway to AI. It can be very difficult to tell whether an image created by a human infringes on a particular copyright, and I think all of us who follow court rulings in this area are sometimes surprised in one direction, sometimes in the other. In general, Commons approach has been to be pretty conservative: we tend not to host a user-made image that we think is at all likely to infringe a copyright; I think we go beyond just "is it more likely than not to infringe" to "would it be reasonable to consider it an infringement". I don't see us having any reason to apply a different standard to AI-generated artwork in terms of judging likely copyright infringements. (Keep in mind: except in the rare case where there has been a legal ruling, these are always judgement calls in terms of what we think a court would say.)

The trickier issue is one of scope. We seem to be forming a consensus that (with the possible exception of images that are specifically hosted as examples of the capability of generative AI) the hurdle for generative AI to be considered in scope is at least as high as that for user-created non-photographic art, and possibly higher. There are a handful of users (Prototyperspective clearly being one of them) who seem to want a more generous scope for generative AI, but I think it's safe to say that among those of us who have been discussion this, at least two thirds of the participants either opt for "same standard as user-created art" or something even stricter. - Jmabel ! talk 23:10, 25 January 2024 (UTC)

Jmabel whilst the copyright status under law is still somewhat unclear, I think ((as you already say) Commons needs to have a position on this matter. The question I have is how to know what that position is? We seem to have some fairly clear guidelines around derived non-generated images, could we come up with guidelines around AI generated images? - Chris.sherlock2 (talk) 00:47, 26 January 2024 (UTC)

@Chris.sherlock2: Since this is on someone else's talk page, I was trying to stay mostly away from things that are just a matter of my own opinion, but as far as works that derive from copyrighted works, and in terms only of possible copyright infringement, I'd ignore the fact that generative AI was involved. As I said above, my guess is that in terms of infringement, I'd be surprised if courts decide that the involvement of generative AI has any bearing. Scope is a separate matter. My own take would probably to treat them the same way we treat user-generated illustrations, which is to say a pretty high bar for being in scope, but not an outright ban. - Jmabel ! talk 02:47, 26 January 2024 (UTC)