A screenshot of the wip, mirrored and flipped, vs post #5825389 look identical. What makes you think it's fake?
Posted under General
A screenshot of the wip, mirrored and flipped, vs post #5825389 look identical. What makes you think it's fake?
nonamethanks said:
A screenshot of the wip, mirrored and flipped, vs post #5825389 look identical. What makes you think it's fake?
I suppose I meant to say the photo they posted, not technically a screenshot. The state of the WIP captured on the photo does not appear at any point in the video, which makes no sense for an unedited CSP timelapse, as it records everything including the hiding/showing of layers. Even ignoring that the video doesn't hold up to scrutiny, eg the sketch phase would have you think the background's hand-drawn but then it appears out of thin air, as if completed in a single stroke.
If you squint really hard, you could pass off the 80 unnamed, ungrouped empty layers and 790px wide canvas as part of the artist's highly unusual personal workflow, but you know what they say about hearing hoofbeats, even more so when they come from an October account that has blatant unedited AI art as their banner and avatar and clearly lied about their previous work.
Updated
Looks an awful lot like an AI art. The resolutions are divisible by 64 (768x1536) and some of the other posts were tagged with AI-related tags.
Edit: Oh, please check post #5866766 too.
World_Funeral said:
Excuse me, do you consider this to be AI-generated despite the tags on Pixiv doesn't say AI at all? It's still mesmerizing and colourful. Sorry if I ask about this before posting in order to avoid deletion instead of asking about existing post(s). Aren't most of the artist's works AI too?
This is 100% AI generated image. They're just being dishonest on pixiv. Their twitter says "AI Engineer/MachineLearning/ComputerVision/AI Art Generation/AI絵をアップロードためのアカウント/"
Jerrpanese said:
Hey yo guys. So when browsing DeviantArt, I came across a user that shared a handy tool website that can help check if an image is AI-generated: https://www.illuminarty.ai/en/illuminate
It's pretty damn useful too. Tested it out on various images, on both AI-generated and authentically drawn content, and the AI-generated stuff always returns high probability of being detected as such by the checker (having an 80-90% probability). I suppose it can also be fun to use this checker to roast AI-generated posters who pretend to be "artists" by showing them these results.
It's not super acccurate but might detect some ai stuff. I've uploaded some of good ai gen images made by same person and one showed 90% probablity and another 9.97%( tho it fooled me as well, maybe it was AI assisted img instead of fully ai). Also real artists can have high % as well. Up to 35% was a common percentage I got for real artists, however, anmis old work got whooping 77%. It's a useful, still developing tool but it's better to check artists account before blindly trusting the website.
VR-Man said:
Looks an awful lot like an AI art. The resolutions are divisible by 64 (768x1536) and some of the other posts were tagged with AI-related tags.
Edit: Oh, please check post #5866766 too.
First post on Pixiv: Oct 15
Most recent one: Dec 3
Number of posts on Pixiv: 45
That's an average of about one piece a day, for one and a half month straight.
It's just too suspicious. I can't trust any account that started uploading in Oct/Nov 2022 - especially this art style - at all.
kind of off-topic but shouldn't this thread be added to the useful thread index?
topic #8235
https://www.pixiv.net/en/artworks/102925065 I ran this through an AI checker before and it said 50% probability. Very helpful.
Nameless_Contributor said:
After playing with this for a bit I don't think it should be used to decide for or against uploading an image to Danbooru.
It seems to give big weight to realistic shading, giving high scores (often above 80%) to many photos or images that would fit in the Expert Shading pool while giving low scores to AI-generated images that don't use common AI artstyles.
Yeah, I'd be wary too. To test this Illuminarty tool, I threw an obvious stitched screencap at it (which is obviously not AI generated and easily verified as authentic via trace.moe), and it comes back with a "confident enough to get scientifically published" confidence of 96.38%. The fact that we're trying to screen out the output of AI models trained on the legitimate art from this very site as input also seems pretty problematic for any sort of algorithm to accurately and confidently segregate the two.
I've pretty much ignored all the debate on AI art here, and I'm perfectly fine with the policies we came up with with regards to it, but how much of a witch hunt do we want to make this? I mean if something is obviously AI generated with deformed hands and limbs etc, obviously it should be not posted, flagged, or left unapproved.
However, If it's basically indistinguishable, is it really worth making false-positive decisions based only on a hunch, or an algorithm's call, or the fact that an artist may have experimented with AI elsewhere or posted something related to AI (but perhaps also legitimately drew things)? Where is the line drawn on AI-assisted being over the line? Artists have always used things like 3D modeling software or photo references to frame something out, and then paint / trace over top of it. If they do the same thing with AI output, but all the strokes and shading are by their hand, is that really that different?
I mean this thread has a place and we should continue ferret out relatively clear cases, but it feels like in some cases we're really pushing to delete things that may not warrant it.
Updated
Yeah, it's getting quite ridiculous. I find it pretty pathetic that now people are advocating for some sort of moral crusade, while at the same time using AI tools to identify AI art because they themselves are incapable of telling it apart from human-made art reliably. If we're going to try to crawl up artists' ass to the point where even a full making-of video from sketch to finished image is not enough to prove that something has been drawn by a human then we might as well just shut down the site, because it's only going to get worse in the coming months (not even a matter of years).
This witch hunt is impossible to win anyways. Even before the AI if you pulled up any two artworks and told me to guess whether they were made by the same artist it would be very hard to guess it correctly. Most artists already learned from each other and very few actually have their own unique style.
And now we have an AI which learned from every artist on the internet and by very definition it draws exactly like most artists. Trying to perfectly separate AI from human works is such an incredibly pointless effort.
If an artist started after October 2022, constantly draws like the AI, and has done nothing to prove their legitimacy, then I would start to get suspicious and if they do use AI you could just blanket ban all of their artworks. But trying to do this individually for every single individual image is a pointless waste of time imo.
I will say this, but if an artist just uses AI the only person who they are cheating is themselves. They will never see their true artistic potential and will always be severely limited by the extreme limitations of only typing keywords into a box. And receiving money and/or fame by cheating with AI when they really don't deserve it, I would just feel incredibly ashamed.
This feels like an unnecessary amount of catastrophizing over one user trying to hide the fact that their art is AI-assisted, when we've been dealing with artists trying to hide their tracing and occasional outright art theft since the site's birth. It's not possible for this site to have perfect moderation, but when has it ever? I'm not seeing where we have made any bad calls so far.
feline_lump said:
This feels like an unnecessary amount of catastrophizing over one user trying to hide the fact that their art is AI-assisted, when we've been dealing with artists trying to hide their tracing and occasional outright art theft since the site's birth. It's not possible for this site to have perfect moderation, but when has it ever? I'm not seeing where we have made any bad calls so far.
You're comparing ai-assisted to tracing, when you should compare it to 3d instead. Stuff like post #5826109, post #5305387, post #5472259 are all obvious tracing over 3d poses, and yet nobody's ever had an issue with any of them (and historically we've even allowed tracing too, unless it was blatant art theft).
Yet in this very topic we reached the point where we're arguing that https://twitter.com/YUA_andI/status/1592298951595937792 is not enough to prove that a post is not ai-generated, and we're getting flags for content like post #5874259, or rejecting post #5868622 when the artist did an entire twitch stream redrawing over it, or counting the layers in their photoshop window to somehow prove it's not a legitimate drawing. Come on.
Updated
nonamethanks said:
You're comparing ai-assisted to tracing, when you should compare it to 3d instead. Stuff like post #5826109, post #5305387, post #5472259 are all obvious tracing over 3d poses, and yet nobody's ever had an issue with any of them (and historically we've even allowed tracing too, unless it was blatant art theft).
It's a matter of the degree to which something was 'copied'. It's one thing to use a blank anatomy model as a reference to create a character with an unique identity, combining it with various elements and concepts of the artist's own imagining, another to paint over a ready-made product almost 1:1 just to obfuscate its place of origin. The latter is at the tail end of the "referencing to tracing" scale.
nonamethanks said:
Yet in this very topic we reached the point where we're arguing that https://twitter.com/YUA_andI/status/1592298951595937792 is not enough to prove that a post is not ai-generated
But that's not true. AI-assisted was the accusation. It was identified as such, tagged accordingly, and left at that, like thousands of other cases of derivative work.
Downright AI generated works should definitely be banned, they're incredibly lazy and low effort, but a full redrawing of an AI work - meh, that isn't that bad. Technically it was drawn by a human at that point. Before AI, many artists just traced over 3D models and copied bits and pieces from reference and I don't really see this being much different.
Again like I said before the only person who they are really cheating is themselves. If they don't want to make anything new and just want to copy the AI then they will never see their true potential. But for danbooru it is not only pointless but will be literally quite impossible to go after redraws of AI works, since even if we get a perfect AI detector for AI works they will pass off as human work since they were redrawn by a human.
Diet_Soda said:
But that's not true. AI-assisted was the accusation. It was identified as such, tagged accordingly, and left at that, like thousands of other cases of derivative work.
But there's not even proof that it's ai-assisted other than another artist being butthurt at them on twitter and throwing accusations without proof. Again, what part of https://twitter.com/YUA_andI/status/1592298951595937792 shows that it's ai-assisted? You can see every single piece being drawn from zero in that making-of. If you take a screenshot from the end of that video, like I did, and overlay it to the full post, they're the exact same picture. What part was generated by an AI? I'm not trying to be smug or make snide remarks, I genuinely cannot see anything that would suggest that, despite multiple people in this thread claiming it's obvious.
Maybe I'm wrong, maybe I'm right for this specific post, but in general we should only label posts as ai assisted or ai generated if there's real, tangible proof of it. Someone complaining on Twitter, the number of layers in Photoshop, none of that is good enough. What do you think is going to happen when a legitimate artist sees some random guy call them a fraud with no proof on these forums?
Updated
nonamethanks said:
But there's not even proof that it's ai-assisted other than another artist being butthurt at them on twitter and throwing accusations without proof. Again, what part of https://twitter.com/YUA_andI/status/1592298951595937792 shows that it's ai-assisted? You can see every single piece being drawn from zero in that making-of. If you take a screenshot from the end of that video, like I did, and overlay it to the full post, they're the exact same picture. What part was generated by an AI? I'm not trying to be smug or make snide remarks, I genuinely cannot see anything that would suggest that, despite multiple people in this thread claiming it's obvious.
Like I pointed out earlier, the video has several parts cut out, so calling it full progress is already disingenuous on their part. The "butthurt artist" is right - Ina WIP is a textbook example of a NAI output and it's bewildering that they decided to post it, anyone who's ever dabbled in drawing can attest that humans, while prone to mistakes, don't just confuse hair with skin and clavicles. Yeah the end result is painted over with a human hand but it was very heavily based off an AI generation, hence AI-assisted.
If the Fauna is drawn in the exact same style, and we've established the artist to be an AI user who has lied and gone out of their way to conceal parts of their process in both pieces, isn't AI-assisted the logical conclusion? Or do we take the microscopic chance they changed their ways overnight, the piece's base properties and composition only coincidentally happen to follow every Novel AI pattern in the book, and the video's missing links are an unfortunate consequence of them accidentally CTRL+V'ing their wife's nudes onto the canvas and keeping them there until they were finished drawing specific elements?
Updated
Jerrpanese said:
Hey yo guys. So when browsing DeviantArt, I came across a user that shared a handy tool website that can help check if an image is AI-generated.It's pretty damn useful too. Tested it out on various images, on both AI-generated and authentically drawn content, and the AI-generated stuff always returns high probability of being detected as such by the checker (having an 80-90% probability). I suppose it can also be fun to use this checker to roast AI-generated posters who pretend to be "artists" by showing them these results.Edit: Oh, it seems like it isn't as useful as it made itself out as. I guess it's better to use our own human judgement.
Big oof from me. Shoulda used my better judgement. ☹
