Is post #7181979 Ai?
Posted under General
Is post #7181979 Ai?
skywalkerred said:
Is post #7181979 Ai?
The lineart looks like typical AI lineart, which could be a coincidence if it weren't for that artist having a bunch of deleted posts that clearly are AI, so I highly doubt this one isn't AI.
Provence said:
Can't agree more with nnt here.
And now we finally have a clear stance on how to deal with ai-assisted. I've been waiting for this for ages. Because either you completely allow it or completely disallow it. You can't make a treshold.
As to why this stuff still gets approved: Accidents or these apiprovers genuinely don't care. In the latter case, it's not valid to allow AI more and more, but instead tell these approvers off (if there is a pattern).
Yeah, but how you can tell these approvers don't care about their task and only approve in a random way? We are humans after all, and how nonamethanks explained, approvers have a tight schedule, dealing with dozens, and perhaps hundreds of images in a single day. What are you going to do with the approvers who unintentionally left a pair of AI images pass the queue? How do you plan to separate those who made a mistake to those who doesn't really care about what they're doing? Subject them to an interrogation is useless, since they can lie, or be honest about the fact they made a mistake approving an AI image. And while you are busy questioning your approvers, the constant flow of uploads continues but the flow of approving descends due to the fact you are retaining your approvers.
And what pattern can be used to know if an approver is neglecting their tasks and allowing AI images? The number of images? How I said before, approvers deal with a big amount of images, so tracking all the AI images they approved will likely be a nightmare. And as the line of human-made art and AI-art is becoming blurrier and blurrier, forcing them to always check the source or seek for proofs of an artist using AI or not is a deficient way to deal with that problem, once again, because of the number of images constanty being uploaded in the site. You are an approver, so you can tell how many images you deal with daily.
SlaughteredMelon said:
Yeah, but how you can tell these approvers don't care about their task and only approve in a random way? We are humans after all, and how nonamethanks explained, approvers have a tight schedule, dealing with dozens, and perhaps hundreds of images in a single day. What are you going to do with the approvers who unintentionally left a pair of AI images pass the queue? How do you plan to separate those who made a mistake to those who doesn't really care about what they're doing? Subject them to an interrogation is useless, since they can lie, or be honest about the fact they made a mistake approving an AI image. And while you are busy questioning your approvers, the constant flow of uploads continues but the flow of approving descends due to the fact you are retaining your approvers.
And what pattern can be used to know if an approver is neglecting their tasks and allowing AI images? The number of images? How I said before, approvers deal with a big amount of images, so tracking all the AI images they approved will likely be a nightmare. And as the line of human-made art and AI-art is becoming blurrier and blurrier, forcing them to always check the source or seek for proofs of an artist using AI or not is a deficient way to deal with that problem, once again, because of the number of images constanty being uploaded in the site. You are an approver, so you can tell how many images you deal with daily.
If an approver consistently approves obvious AI, then it can be treated in the same way as when an approver consistently approves obviously low-quality posts: by brining it up, DMail, feedbacks, etc.
Updated
岩戸鈴芽 said:
If an approver consistently approves obvious AI, then it can be treated in the same way as when an approver consistently approves obviously low-quality posts, by brining it up, DMail, feedbacks, etc.
I see. Thanks for bringing some insight to this matter. The feedback of an approver in this matter helps.
Provence said:
Can't agree more with nnt here.
And now we finally have a clear stance on how to deal with ai-assisted. I've been waiting for this for ages. Because either you completely allow it or completely disallow it. You can't make a treshold.
As to why this stuff still gets approved: Accidents or these apiprovers genuinely don't care. In the latter case, it's not valid to allow AI more and more, but instead tell these approvers off (if there is a pattern).
While there is a clear stance (though I'm way more confused than when I thought all AI was banned), the ai-assisted tag is an absolute minefield. Looking through it and seeing what gets past and what gets flagged, I don't understand what's acceptable and what's not. Some AI features are obvious to me and get through. Others aren't and are the cause for getting flagged and deleted. I understand some things people point out. Other times, I can't see the flaws even after the people with Mystic Eyes of AI Perception point them out.
I've tested holding ai-assisted art to the same standard as other art on this site. The results are far from inspiring, but I can say I've learned some beneficial things that would hopefully make me a better approver in the future. That said, I don't really want to touch the ai-assisted tag with a 10-foot pole right now. I cannot tell to what degree flags are actually justified, but because I take flags seriously, I have to assume they are completely justified.
VR-Man said:
See comment #2390470.
I don't see any obvious AI artifacts, and the fusing hair mentioned in the comments is a far stretch for claiming it's 100% ai-generated. It could very well be ai-assisted, but even so, I personally don't see anything worth rejecting/deleting it over in that case.
Can we reinstate to manually delete AI-Gen "art" if they reach the 1st page in the popular tab?
Just deleted two of such uploads. And while I'm sorry to nuke an uploaders upload slots, it just looks bad. Or else we might as well discuss the unbanning of high-scoring AI gen images, which btw. would only coaintain BA and GI, so it'd be copyright-dependant. I don't see the point to keep these active or pending.
Updated
Provence said:
Can we reinstate to manually delete AI-Gen "art" if they reach the 1st page in the popular tab?
Just deleted two of such uploads. And while I'm sorry to nuke an uploaders upload slots, it just looks bad. Or else we might as well discuss the unbanning of high-scoring AI gen images, which btw. would only coaintain BA and GI, so it'd be copyright-dependant. I don't see the point to keep these active or pending.
Both of those uploads are ai-assisted. Even both artists' bios say they're assisted.
nonamethanks said:
Both of those uploads are ai-assisted. Even both artists' bios say they're assisted.
So, it means we should flag ai-gen AND ai-assisted?
nonamethanks said:
Both of those uploads are ai-assisted. Even both artists' bios say they're assisted.
I totally get we can't afford to spend fifty posts discussing minute details of every image, but something like post #7232472 that already looks awful from the thumbnail and assaults your eyes with this bullshit the moment you open it getting away with "assisted" is confusing, even under very lax AI gen rulings.
Diet_Soda said:
I totally get we can't afford to spend fifty posts discussing minute details of every image, but something like post #7232472 that already looks awful from the thumbnail and assaults your eyes with this bullshit the moment you open it getting away with "assisted" is confusing, even under very lax AI gen rulings.
Almost all of their images have multiple different versions (paywalled), both nude and clothed, which clearly shows that, at the very least, the images went through significant editing. Sure, we can argue that maybe all they do is just copypaste nipples or bras on top of the variants. But it's still not fully ai-generated content.
I undeleted that post that Provence mentioned because it was directly deleted when it shouldn't have been, not because I particularly liked it. If you believe it's of poor quality then flag it and let the queue decide.
Updated
nonamethanks said:
Almost all of their images have multiple different versions (paywalled), both nude and clothed, which clearly shows that, at the very least, the images went through significant editing. Sure, we can argue that maybe all they do is just copypaste nipples or bras on top of the variants. But it's still not fully ai-generated content.
You can create entire variant sets in a minute via inpainting or reusing seeds, and seeing how lazy their base images are there is no doubt the others suffer from the same issues. I still strongly feel that assisted should be reserved for "a modicum of effort was put into making this image not look like a garbled mess at a glance" because otherwise this kind of thing will quickly start to clog the queue when it doesn't really need to.
One of my uploads has been flagged - 7223997
I believed it was normal because the creator showed progress with the final image in the comments - https://twitter.com/ARMSWhite/status/1753757646300262538
However, they do mention digital correction. Is there a threshold that has been overstepped here?
Updated
WhiskyWin said:
One of my uploads has been flagged - post #7223997
I believed it was normal because the creator showed progress with the final image - https://twitter.com/ARMSWhite/status/11753758667185463658
However, they do mention digital correction. Is there a threshold that has been overstepped here?
It looks like this artist may have used img2img on a commission they received from someone else. I can't read the signature on the original sketch, but it looks like a different name. The AI features pointed out in the flag are not present in the original sketch(es). Definitely ai-assisted to some degree imo.
