Is this even ai generated? I'm usually really good at spotting those since this I pay close attention to images. Can't find any of the usual mistakes or inconsistencies ai makes Couldn't find any mention of ai on the artist page as well so, why exactly?
Is this even ai generated? I'm usually really good at spotting those since this I pay close attention to images. Can't find any of the usual mistakes or inconsistencies ai makes Couldn't find any mention of ai on the artist page as well so, why exactly?
While it doesn't necessarily imply AI, the fact that they're pumping out daily, sometimes more than 1 image a day, of similar style images is extremely suspicious. Yes, some artists can just do that, but it's prettty damn suspicious.
While it doesn't necessarily imply AI, the fact that they're pumping out daily, sometimes more than 1 image a day, of similar style images is extremely suspicious. Yes, some artists can just do that, but it's prettty damn suspicious.
Also this has the common AI resolution (832x1216).
In the moderation history for post #10081349, the flagger gives the reason "AI (see artist's Twitter)", as if looking at it will make it obvious, but it doesn't to me.
There are two things I did notice: firstly, his output rate is very high, as @X-Sam pointed out
The second is highly subjective. I've noticed there is a tendency for the work of AI artists to be "oddly specific". For example, one of the first series of AI posts here featured Nagatoro Hayase dressed in ancient egyptian clothes: see post #5747284 et al. Although not currently tagged as ai-generated, I have strong suspicions regarding the artist tumugihikari, in part due to the repeating theme of "Touhou characters wearing bikinis smirking at the viewer, in greyscale with colored background and with spot colored eyes".
In the moderation history for post #10081349, the flagger gives the reason "AI (see artist's Twitter)", as if looking at it will make it obvious, but it doesn't to me.
There are two things I did notice: firstly, his output rate is very high, as @X-Sam pointed out
The second is highly subjective. I've noticed there is a tendency for the work of AI artists to be "oddly specific". For example, one of the first series of AI posts here featured Nagatoro Hayase dressed in ancient egyptian clothes: see post #5747284 et al. Although not currently tagged as ai-generated, I have strong suspicions regarding the artist tumugihikari, in part due to the repeating theme of "Touhou characters wearing bikinis smirking at the viewer, in greyscale with colored background and with spot colored eyes".
At the very least the SFX are very unlikely to be AI generated.
It's also probable that it's ai-assisted, which explains the fast turnout but retains the real quality. Because I looked this image over and over to put 70 tags on it I did not find a single sign of ai. Wouldn't upload it otherwise
Yeah, I took a bigger gander at it. Went over the exif. Checked every mention of the artist, scrolled for a lot of time to look at his works. Compared it and other images to different models, and the only thing I could find that there's ai is the high output of images, the repetition of motifs and the resolution. This makes it obvious the ai is involved in the process but the images are just too well made and too different from ai models to be fully ai generated. And I checked many of them
The first thing I check when suspecting an image of being AI-generated is the eyes, artists usually like to keep them symmetrical. The upper part of the irises here aren't consistent, and the upper eyelash of the right eye blends into her hair. One of the right eye's lower eyelash is also white for no reason. There are also other things that stick out, the black strap of her dress on the right doesn't appear near her armpit (it blends in with her hair) I know this is extremely nit-picky but I'm leaning towards full on AI.
In the moderation history for post #10081349, the flagger gives the reason "AI (see artist's Twitter)", as if looking at it will make it obvious, but it doesn't to me.
I'm the one who flagged it and the reasons you give are the reasons why. If you scroll their media tab there are lots of things that should set off your AI warning bells:
High output volume.
Repeatedly drawing the same thing with different characters (a sign of reusing prompts).
Everything is 832x1216 (NovelAI's default resolution) or upscaled from 832x1216 (because they can't do any other resolutions; notice how blurry post #10083821 is because it's upscaled from 832x1216).
Oldest work has the characteristic NovelAI look and lighting style (most models are heavily trained on photos, which bleeds into art having an uncanny photographic quality, where everything looks like a brightly lit photo instead of a less realistic anime drawing).
Style change after March 2025, when NovelAI v4 came out (compare the lighting on their newest works versus their oldest works; notice how it's softer and more subdued and has less of that bright photoreal quality).
Inability to draw the same character the same way twice (compare [1], [2], [3], [4]; notice how the details of the clothing and accessories are off-model and never drawn the same way twice).
No signatures (humans tend to sign their work, especially these days; AI artists tend not to, either because they have no reason to take pride in their work or because they can't even draw a signature).
No Twitter activity besides pumping out art. Human artists tend to talk to people, follow other artists, post about things other than their own art, or be involved in some kind of fandom; AI artists (unless they're grifting a Patreon) have a tendency to keep to themselves and do nothing but dump AI gens, possibly to avoid getting called out.
This is all circumstantial evidence, but taken together it paints a picture. Looking for minor flaws in the image is a meme, most models these days are good enough that there are no clear flaws or giveaways any more.
I'm the one who flagged it and the reasons you give are the reasons why. If you scroll their media tab there are lots of things that should set off your AI warning bells:
High output volume.
Repeatedly drawing the same thing with different characters (a sign of reusing prompts).
Everything is 832x1216 (NovelAI's default resolution) or upscaled from 832x1216 (because they can't do any other resolutions; notice how blurry post #10083821 is because it's upscaled from 832x1216).
Oldest work has the characteristic NovelAI look and lighting style (most models are heavily trained on photos, which bleeds into art having an uncanny photographic quality, where everything looks like a brightly lit photo instead of a less realistic anime drawing).
Style change after March 2025, when NovelAI v4 came out (compare the lighting on their newest works versus their oldest works; notice how it's softer and more subdued and has less of that bright photoreal quality).
Inability to draw the same character the same way twice (compare [1], [2], [3], [4]; notice how the details of the clothing and accessories are off-model and never drawn the same way twice).
No signatures (humans tend to sign their work, especially these days; AI artists tend not to, either because they have no reason to take pride in their work or because they can't even draw a signature).
No Twitter activity besides pumping out art. Human artists tend to talk to people, follow other artists, post about things other than their own art, or be involved in some kind of fandom; AI artists (unless they're grifting a Patreon) have a tendency to keep to themselves and do nothing but dump AI gens, possibly to avoid getting called out.
This is all circumstantial evidence, but taken together it paints a picture. Looking for minor flaws in the image is a meme, most models these days are good enough that there are no clear flaws or giveaways any more.
Interesting So the way to find ai now is to look at the bigger picture?
Looking for minor flaws in the image is a meme, most models these days are good enough that there are no clear flaws or giveaways any more.
I disagree, the eyes are rather obvious here still and immidiately gave it away for me. Those are way too smudgy and inconsistent, the supposed sparkles are more like crosses or X. The right iris also bleeds into the lower lashes.
The jewelry(?) on her upper chest is another semi-obvious tell as its smudgy and asymmetrical. Handful other accesories look off too and there is this overall lack of ''sharpness'' when it comes to various details (round edges when they should be pointy).
I'd say that it is still very much possible to tell it's AI from details but I do admit that if proficiently touched up, this would be very hard to tell its AI. Though at that point they would be have makings of an proper artist.
Leave a comment