post #9767040 somebody is asking if this is ai in the comments; I don’t think it is, but I’m not certain.
I'm not certain either and as comment #2537241 mentioned the hair's shadow does not match the hair (I'm not convinced by the other points). kyatsuwu's shading tends to have a single-hard-light, low-diffusion style, which causes high contrast and sharp transition between midtone and shadow, and on top of that the border between midtone and shadow have a strip of high saturation, pink-orange band. Both things coincide with the "AI vibe", so I'm not surprised that this artist attracts suspicion, or if they turn out to be painting over AI (ai-assisted). For now I'd say the proof is insufficient, but this artist should be kept an eye on.
Calling out an established artist in particular for ai requires way more than just "a vibe" tbh. Checking for available good time lapses and psd's should be the bare minimum. Also not conflating common artist mistakes as ai on its own.
For a lot of these artists, its their livelihood. A false ai accusation can literally be devastating for them so bringing forth strong evidence should be the standard. It reflects badly on the site to be wrong about things like this imo.
This. This is my point. When not befuddled myself, I am here to flag the blatant posts that slipped past the radar, and having them reported here is supposed to not be a terrible idea. The fact these AIs are usually trained on the popular styles too means artist who've been drawing that way for years are going to get accused left and right. Playing the blame game like this also distracts heavily from the legit questions and catching actual prompters, even ignoring the fact a false accusation can be utterly devastating to an artist...
While I don't really know how enforceable it'd be without admin input, I would propose these guidelines as a starting point for accusing established artists:
1. Checking for psd files and timelapses should be the bare fucking minimum, as NiceLittleDan said above. 2. Generally speaking, accusing established artists requires being good at spotting AI to begin with. And by that, I mean spotting AI artifacts and noise, and possibly legit nonsense (like melting backgrounds, not a mishapened hand any artist can fuck up). You also should ideally have an extensive catalogue of such errors...Though if you have that sort of time and energy, why start your campaign on Danbooru, I ask? 3. Actually suspicious behavior or perhaps even a fucking Twitter expose thread if you can verify it's not mere conjecture.
Basically, all of this is to say "You should do your own investigation if you care that fucking much about it". AI is not a light accusation at all, and we should really start actively treating it as such when levied against established artists. Or anyone not blatantly advertising AI slop really, but established artists have the most to lose by a false accusation.
No matter what though, we really need to discourage witchhunting here. I think making it actively harder to accuse established artists based on mere vibe is a good starting point at least, no?
I asked for a list about what "evidence" is allowed to show an artwork is AI-generated in this thread two months ago but nobody gave me any answer. The "hey I am not gonna tell you what is not allowed but what you are doing is not allowed" game is still going on huh.
There will never be any concrete answer to this question because the technology changes over time, but the campouts in this thread who contribute nothing to the site except accusing random artists of using AI need to clearly do far more than "vibe check" or conclude that completely human-borne errors are marks of AI generation. However, when it's clear that you don't care beyond a surface inspection and start pointing fingers randomly, then yes, that shit is going to get on someone's nerves, which you and ai-tracker are successfully doing.
Do not antagonise artists. We don't need random people going on a vigilantism spree and harassing artists because they saw an unsubstantiated comment or flag on Danbooru saying they do use AI.
There will never be any concrete answer to this question because the technology changes over time, but the campouts in this thread who contribute nothing to the site except accusing random artists of using AI need to clearly do far more than "vibe check" or conclude that completely human-borne errors are marks of AI generation. However, when it's clear that you don't care beyond a surface inspection and start pointing fingers randomly, then yes, that shit is going to get on someone's nerves, which you and ai-tracker are successfully doing.
Do not antagonise artists. We don't need random people going on a vigilantism spree and harassing artists because they saw an unsubstantiated comment or flag on Danbooru saying they do use AI.
Just in case anyone missed or forgot, araizumi_rui's post #6300289 got tagged as ai-generated and the unlucky guy in the comments got 7 downvotes for saying "this is not AI-generated and actually the artist drawing it like this".
If there is any reason other than the "vibe" for tagging it as ai-generated, I am glad to hear that. Otherwise, I would recommend that you remove this tag, dear site contributor.
For one, that post was flagged two years ago, and I know you have a very specific grudge against me right now, so in case it isn't obvious I wasn't around then. For two, if you would take five seconds of your day and check the artist's tag you would see that they have currently active posts.
What is the point in giving you the option to appeal if you won't do anything but complain about the judgement of posts - which, every post is judged on its individual merits - or expect that others do either the investigative work for you or stick up for cases that feel like a misjudgement? Appeal it if you think it's an error.
For one, that post was flagged two years ago, and I know you have a very specific grudge against me right now, so in case it isn't obvious I wasn't around then. For two, if you would take five seconds of your day and check the artist's tag you would see that they have currently active posts.
What is the point in giving you the option to appeal if you won't do anything but complain about the judgement of posts - which, every post is judged on its individual merits - or expect that others do either the investigative work for you or stick up for cases that feel like a misjudgement? Appeal it if you think it's an error.
Well, actually, I am not really expecting you to do any work for me. Did you forget that we are playing the "hey I am not gonna tell you what is not allowed but what you are doing is not allowed" game? I was just waiting for your saying "hey I am not gonna tell you why it is ai-generated but it is ai-generated" again.
Anyway, it is fun to see people do nothing about the ai-generated tag on some random post of an artist with no reason while saying how much they love artists.
it is fun to see people do nothing about the ai-generated tag on some random post of an artist with no reason while saying how much they love artists.
Including yourself, per WRS's point. At this point, I don't think this discussion has any merit or anything to do with the topic of checking for AI-generation and should probably cease.
Are you just maliciously ignorant? Don't tell me that you aren't expecting others to go to bat for or against potential AI involvement in a post and then sit here showing no signs of actually trying to rectify possible problems and shit up topics that are meant for important discussion. Have the self-awareness to realise that you are part of the "people doing nothing" group you're trying to antagonise. You have both an edit button and an appeal button. Either use them the way they're intended to be used or don't complain.
I can't, for the life of me, even fathom how you think any of this is a game. It's a sick and actively damaging way of thinking. This topic is for discussion; you can refute opinions here and provide alternative lines of reasoning or question someone's judgement. Your snark and antagonism, which you appear to prefer more than good faith discussion, is unhelpful here.
Are you just maliciously ignorant? Don't tell me that you aren't expecting others to go to bat for or against potential AI involvement in a post and then sit here showing no signs of actually trying to rectify possible problems and shit up topics that are meant for important discussion. Have the self-awareness to realise that you are part of the "people doing nothing" group you're trying to antagonise. You have both an edit button and an appeal button. Either use them the way they're intended to be used or don't complain.
I can't, for the life of me, even fathom how you think any of this is a game. It's a sick and actively damaging way of thinking. This topic is for discussion; you can refute opinions here and provide alternative lines of reasoning or question someone's judgement. Your snark and antagonism, which you appear to prefer more than good faith discussion, is unhelpful here.
How about spending some time checking what you said: What "evidence" is allowed to show an artwork is AI-generated in this thread? -> "No any concrete answer" Why post #6300289 is tagged as ai-generated? -> "I don't know why. Do it yourself."
And then here we go again: "I just said nothing helpful but your snark and antagonism is unhelpful"
Anyway, still no answer to the question "what 'evidence' is allowed to show an artwork is AI-generated in this thread" and still nobody is gonna do anything about post #6300289. If everyone thinks that's fine, then that's fine. Have a good day.
So it's malicious ignorance, running circles around the room thinking you've made a solid point. Again I remind you that this post is from 2 years ago and is completely unrelated to me. I did not tag it, nor do I have anything to say in support of or against the tag, but clearly you do, so you ought to do something about it if you feel this strongly about it. I don't, therefore I have nothing to say about that post.
Don't expect other people to do the work for you if you just want to sit here and soapbox about how everyone else is wrong. It's not rocket science mate. Make a case about why you think it is or isn't generated, or use your handy edit, flag and appeal buttons if you think you have a solid case for either side. This topic is for second opinions; it still primarily all comes down to someone's judgement to tag, flag or appeal. You're not committing to any action, instead expecting someone else to do it, and chirping about how "no one will do anything" without any self-awareness that all you do is sit here in this topic and whine instead of doing anything meaningful.
So it's malicious ignorance, running circles around the room thinking you've made a solid point. Again I remind you that this post is from 2 years ago and is completely unrelated to me. I did not tag it, nor do I have anything to say in support of or against the tag, but clearly you do, so you ought to do something about it if you feel this strongly about it. I don't, therefore I have nothing to say about that post.
Don't expect other people to do the work for you if you just want to sit here and soapbox about how everyone else is wrong. It's not rocket science mate. Make a case about why you think it is or isn't generated, or use your handy edit, flag and appeal buttons if you think you have a solid case for either side. This topic is for second opinions; it still primarily all comes down to someone's judgement to tag, flag or appeal. You're not committing to any action, instead expecting someone else to do it, and chirping about how "no one will do anything" without any self-awareness that all you do is sit here in this topic and whine instead of doing anything meaningful.
So you think your proving my malicious ignorance now is something helpful or meaningful? If so, do whatever you want.
How about spending some time checking what you said: What "evidence" is allowed to show an artwork is AI-generated in this thread? -> "No any concrete answer" Why post #6300289 is tagged as ai-generated? -> "I don't know why. Do it yourself."
And then here we go again: "I just said nothing helpful but your snark and antagonism is unhelpful"
Anyway, still no answer to the question "what 'evidence' is allowed to show an artwork is AI-generated in this thread" and still nobody is gonna do anything about post #6300289. If everyone thinks that's fine, then that's fine. Have a good day.
As the person who first brought up post #6300289 in this thread and tagged it as AI-generated, I would ask what's your justification for this weird object on the couch? https://files.catbox.moe/3gmryp.jpg Or the fact that the gold at the top of the couch changes color oddly? https://files.catbox.moe/jl63dm.jpg Or for whatever the fuck is going on with these buttons? https://files.catbox.moe/ils8nz.jpg Or even just the fact that no other art from this artist has this style? post #8068348, post #8069746post #8474167 and post #8931209 are all great examples of Araizumi Rui's actual modern artstyle and not a single one of them look like post #6300289. About half of the replies on the original Twitter post were asking if it was AI or pointing out artifacts as well so it's certainly not a Danbooru-exclusive take that this image was at least partially AI-generated. Many justifications, including mine, were weaker in the early days of the thread but that's because AI was much more primitive and much more obvious. AI has gotten much better now so unless you have actual proof, running around screaming that this or that is or isn't AI just makes you look like an ass.
As the person who first brought up post #6300289 in this thread and tagged it as AI-generated, I would ask what's your justification for this weird object on the couch? https://files.catbox.moe/3gmryp.jpg Or the fact that the gold at the top of the couch changes color oddly? https://files.catbox.moe/jl63dm.jpg Or for whatever the fuck is going on with these buttons? https://files.catbox.moe/ils8nz.jpg Or even just the fact that no other art from this artist has this style? post #8068348, post #8069746post #8474167 and post #8931209 are all great examples of Araizumi Rui's actual modern artstyle and not a single one of them look like post #6300289. About half of the replies on the original Twitter post were asking if it was AI or pointing out artifacts as well so it's certainly not a Danbooru-exclusive take that this image was at least partially AI-generated. Many justifications, including mine, were weaker in the early days of the thread but that's because AI was much more primitive and much more obvious. AI has gotten much better now so unless you have actual proof, running around screaming that this or that is or isn't AI just makes you look like an ass.
Here we go again: you say actual proof, so tell me what is actual proof to show an artwork is AI. Asking for actual proof while never explaining just makes you look like an environmentalist gluing your hands to the ground and asking people to protect the earth.
I would recommend that you remove this tag, dear site contributor.
For the record all users, not just those above certain levels, can edit post tags, and appeal a post to get it reconsidered by approvers. I haven't seen post #6300289 get appealed by anybody yet.
Tools like layerdivider mean showing the layers of a drawing has become less convincing than when generative AI just started being a problem.
This video came out during the Araizumi Rui controversy. It showed the process of creating a plausible-looking CSP file from an AI image using layerdivider.
eromelon said:
I asked for a list about what "evidence" is allowed to show an artwork is AI-generated
This is worth discussing. But before that I need to comment that this thread is just a discussion thread and is not actually part of Danbooru's content moderation process at all. What people in this thread say about a post doesn't lead to changes. To delete a post, somebody still needs to go and flag it, and it still needs to be reviewed by approvers for 72h and only then deleted if no approver re-approves it. Likewise, to get a post back, somebody still needs to appeal it and it needs to be approved within 72h.
I'd like to use post #6300289 to explain what indicators I look for.
Inconsistency across how repeated elements are drawn. For a human artist each repetition should be done in similar ways.
Frills. They're hard to draw, but a good human artist would be able to make the sizes and shapes of folds consistent, and Araizumi Rui certainly is one: post #2828220, post #3012854. AI struggles to do so.
Shoes. Her right-hand shoe has very vague shape, which is not consistent with how detailed the drawing is. Her left-hand shoe should not be invisible given where her leg and hand are.
Hair, eyebrows, and lashes. AI images may have hair and its shadow not matching, hair blending with eyebrows and lashes, or hair and other facial features' linework entangled.
AI linework can have brush width variations that don't make sense. For example when drawing fingers it's not natural to use extra thick lines like the example below. AI can also make the line color vary from light to dark in strange ways. For a human artist, they may use lighter lines for brighter areas and darker lines for places in shadow, but AI does not follow this.
From an earlier comment: single-hard-light, low-diffusion shading, which causes high contrast and sharp transition between midtone and shadow; the border between midtone and shadow have a strip of high saturation, pink-orange band. This is what AI is more likely to create, but not impossible for human to draw.
Tools like layerdivider mean showing the layers of a drawing has become less convincing than when generative AI just started being a problem.
This video came out during the Araizumi Rui controversy. It showed the process of creating a plausible-looking CSP file from an AI image using layerdivider.
This is worth discussing. But before that I need to comment that this thread is just a discussion thread and is not actually part of Danbooru's content moderation process at all. What people in this thread say about a post doesn't lead to changes. To delete a post, somebody still needs to go and flag it, and it still needs to be reviewed by approvers for 72h and only then deleted if no approver re-approves it. Likewise, to get a post back, somebody still needs to appeal it and it needs to be approved within 72h.
I'd like to use post #6300289 to explain what indicators I look for.
Inconsistency across how repeated elements are drawn. For a human artist each repetition should be done in similar ways.
Frills. They're hard to draw, but a good human artist would be able to make the sizes and shapes of folds consistent, and Araizumi Rui certainly is one: post #2828220, post #3012854. AI struggles to do so.
Shoes. Her right-hand shoe has very vague shape, which is not consistent with how detailed the drawing is. Her left-hand shoe should not be invisible given where her leg and hand are.
Hair, eyebrows, and lashes. AI images may have hair and its shadow not matching, hair blending with eyebrows and lashes, or hair and other facial features' linework entangled.
AI linework can have brush width variations that don't make sense. For example when drawing fingers it's not natural to use extra thick lines like the example below. AI can also make the line color vary from light to dark in strange ways. For a human artist, they may use lighter lines for brighter areas and darker lines for places in shadow, but AI does not follow this.
From an earlier comment: single-hard-light, low-diffusion shading, which causes high contrast and sharp transition between midtone and shadow; the border between midtone and shadow have a strip of high saturation, pink-orange band. This is what AI is more likely to create, but not impossible for human to draw.
I appreciate your patience and good faith but this is actually a trap. If anyone literally gives me a list about what "evidence" is allowed to show an artwork is AI-generated, then it would be used to judge every discussion in this thread in the future. Like, when anyone says something is not AI-generated because of the CSP file, I would run to their face and say "dude we got layerdivider now so this is not reliable if you disagree then tell me what CSP file is reliable so I can choose next target based on your explanation ha ha". That's why smart contributors choose to say nothing about this "list" while asking for actual proof at the same time.
So obviously the only choice for artists protecters is to talk about why they don't think the artwork is AI-generated every single time without repeating "actual proof" "actual proof" again and again. However, people just enjoy shutting others' mouths with one single word "actual proof". I am not sure why they don't think AI witch hunters would find this is just another hunting game and join the new game, but if they enjoy it, enjoy it.
Goodness FUCKING gracious dude, several people have tried to engage you in good faith and asked you to stop acting like an asshole and you're just fucking dragging this out for no reason.
If you think it's AI-generated then flag it and do the bare minimum to provide proof or point out the problems.
If you think a post was misjudged, then appeal it and counter the flag reason.
Every post is judged on its own merits. There will never be one universal answer because every post is different. Approvers make the final call.
What is hard to understand about that? What is the point of being obstinate like this? You have done literally nothing else other than chase your own tail around the room with things I've said to you, and when the person who did actually do something in regards to the post you were complaining about replied with why they did so, that too wasn't enough for you.
This post is purely for actual discussion and second opinions. Nothing here ever holds any weight. No word said here is gospel, fact or anything of the sort. You don't have to consider other people's opinions here to take action, you can even do a fun thing called ignoring people, and thinking that this thread is anything more than that is outright wrong.
If you continue shitting up the thread I'm just going to give you a negative for contributing nothing other than bad faith flame wars and trolling on the forum. It's time to move the fuck on. It's clear we can't get you to give a damn to actually do anything useful except complain and argue with people who think you're either wrong or that you're being an asshole to others for nothing other than to grandstand some self-righteous anti-AI stance.
