post #8309517 was flagged as AI. Artist posted very clearly non-AI in the past. This new set of artworks looks slightly different but it's not an unbelievable improvement after a year and doesn't look much like AI to me.
Posted under General
post #8309517 was flagged as AI. Artist posted very clearly non-AI in the past. This new set of artworks looks slightly different but it's not an unbelievable improvement after a year and doesn't look much like AI to me.
ion288 said:
The sword is suspicious, especially the text/inscription. I don't know much about Fate, but that sword seems to be clarent (fate), and looking at other artwork of it, the design looks off in this one. It's like, the same basic shape and color scheme but most of the details are different. This from type-moon wiki is what it probably should look like, and most art here looks a lot more like that.
Also some questionable things going on where hair meets eyes, which is one of the first places I look for AI.
And I guess it could be the angle, but... she doesn't appear to have thumbs? The fingers on the hand toward the end of the hilt are kind of blurred together too.
I want to bring up artist #289727 tor_(torkrub111) again.
They've come up at least thrice in this thread (forum #253272, forum #255334, forum #306436 / comments on post #7831759) and looked suspicious to several people. I (successfully) flagged one of their posts (post #6551675) a year ago where I saw strong evidence of straight-up AI-gen. (I realized just now I'd neglected to tag it at the time, which I've now done. Also, for whatever it's worth, that post is one of two of theirs uploaded here that has since been deleted from pixiv.) In general, however, there was some uncertainty about the level of AI usage and nothing else was tagged or flagged. I feel there is a consensus that they use AI, but nobody has been sure enough of the precise situation to go and take action about it.
Meanwhile their stuff keeps getting uploaded, and in discussion a month ago on post #7831759 the body of already-approved posts became an argument against AI involvement. That troubles me, because I'm personally quite confident that they're using assistance at least.
Additional developments:
How do we handle this?
I've half a mind to mass-tag their posts as ai-assisted at this point. Some of them I think are barely even assisted, but it seems like at least some of them have human work put in, and I'd rather give the benefit of the doubt than condemn someone who's actually drawing. I am pretty sure they're using AI significantly somewhere in their process, though.
iconcawaiine has had four posts flagged but only one of them is tagged. Please tagg them if you feel certain they are AI or I will reapprove them in a day or so.
ion288 said:
iconcawaiine has had four posts flagged but only one of them is tagged. Please tagg them if you feel certain they are AI or I will reapprove them in a day or so.
Please don't reapprove stuff just because it doesn't have a tag. Not everyone is good at spotting AI, but that's not an excuse to go around blindly rejecting potentially valid flags. Hell, some of the disinterests you see may be other approvers approvers thinking "that's too AI-y".
Demanding people add ai-assisted or ai-generated makes it impossible to say "maybe, can someone check", but instead forces it to be a "this is AI, prove me wrong", which is just a bad idea all around.
ion288 said:
iconcawaiine has had four posts flagged but only one of them is tagged. Please tagg them if you feel certain they are AI or I will reapprove them in a day or so.
have you ever taken a moment to consider that adding the tags would compromise the flagger's anonymity?
"sometimes used AI" according to X profile but seems like a pure AI account to me. Trained on sasamori_tomoe, at least partially.
Updated
Didn't want to flag it without bringing it to discussion first, but I get a rather strong AI impression from post #8272208. Just going down my reasoning:
I could be off-base, because the artist has done other work that doesn't give off the same AI vibes (to me, at least - I easily could be wrong there). Even so, I feel rather confident that at least this piece specifically is AI-generated, or at the very least very heavily AI-assisted.
That to me looks like just a mirrored armband (the text and logo are reversed as if looking in a mirror), an artistic error and not really caring for fine details.
Minitajfun said:
yeah this is AI, the model is clearly very good at drawing the patterns on the sweater, however they break down around the sleeves and edges. however the biggest giveaway is the eyes, hair, ear.
not to mention the artist's twitter banner
ANON_TOKYO said:
Please don't reapprove stuff just because it doesn't have a tag. Not everyone is good at spotting AI, but that's not an excuse to go around blindly rejecting potentially valid flags. Hell, some of the disinterests you see may be other approvers approvers thinking "that's too AI-y".
Demanding people add ai-assisted or ai-generated makes it impossible to say "maybe, can someone check", but instead forces it to be a "this is AI, prove me wrong", which is just a bad idea all around.
I said I would reapprove them after people here had a chance to say if they thought they where AI or not. I think its important that AI images gett tagged as such, both to prevent appeals / reapprovals and as a (weak) deterrence against frivolous flags. I would hope that approvers that think an image looks AI-ish would click the Breaks rules button. The images I asked about had several Disinterests but no Break rules.
I should note that Im not the one who reapproved the posts in question.
bipface said:
have you ever taken a moment to consider that adding the tags would compromise the flagger's anonymity?
I have. Sorry if it read as me demanding that the flagger should tag them. I was only asking if someone with better eyes than me could pass judgment.
ion288 said:
The images I asked about had several Disinterests but no Break rules.
When I see a posts that looks a bit too AI-y for my liking, but I'm not sure about it, I won't be pressing "break rules", but I also won't approve it.
I'm probably not the only one who goes about it in such a fashion, so it could very well be possible that every single one of the disinterests thinks it's either ambiguous or too AI. You should leave the reapproving to people who are at least somewhat confident it's not AI, instead of just blindly reapproving just because people haven't explicitly told you what ty think.
ANON_TOKYO said:
Please don't reapprove stuff just because it doesn't have a tag. Not everyone is good at spotting AI, but that's not an excuse to go around blindly rejecting potentially valid flags.
No, the burden of proof is on the flagger. If someone flags a post for "ai-generated" without providing proof, and the post doesn't look like AI slop, then an approver can also approve it without proof.
bipface said:
have you ever taken a moment to consider that adding the tags would compromise the flagger's anonymity?
They can provide proof in the flag reason, otherwise sucks to be them.
This tweet https://x.com/urabiyori/status/1848022818388316209 brought up the possibility of chikoku having AI arts. Therefore, I think this warrants some more thorough AI checking.
First post for AI checking: post #8275569
gzb said:
This tweet https://x.com/urabiyori/status/1848022818388316209 brought up the possibility of chikoku having AI arts. Therefore, I think this warrants some more thorough AI checking.
First post for AI checking: post #8275569
post #6338333 BORN.TO.KILL looks like gibberish, plus inconsistent crosses on backpack. Looks like bad AI copy of post #6185771 or post #4628304.
Marlor said:
It was posted in 2013 on pixiv, I doubt it can be AI back then.
post #8303144 is not the same image as the one on pixiv. Just look at the wonky keyboard. Looks like an AI upscale.
https://danbooru.donmai.us/artists/369587 changed style several times, innormous frequency, noisy filter
Before I upload, can someone give it the AI check.